# Mardi Gras Developers > AI solutions on your own data, hosted securely in the EU. Mardi Gras Developers is a software team of two senior engineers from Breda, the Netherlands. We build AI solutions and private AI platforms, and what lies underneath them: secure infrastructure, data migrations and backends in Kotlin and Java. You work directly with the engineers who build it. - **35,000** active users of our apps in the 2026 season - **7** apps live from one codebase - **6** admin portals in production - **10+ years** of Java, Kotlin and cloud experience In production at, among others: Kielegat, Kruikenstad, Oeteldonk, Boemeldonck, Moesland, Carnaval in Uje, Carnavalsmaatje. ## Seven services around AI, data and backends - [AI solutions and agents](https://mg-devs.nl/en/services/ai-solutions/): Assistants on your own data and agents that complete tasks, measurable and under control. - [Private AI platform](https://mg-devs.nl/en/services/private-ai-platform/): ChatGPT-style convenience for your staff, without company data leaving Europe. - [AI infrastructure and security](https://mg-devs.nl/en/services/ai-infrastructure-security/): Infrastructure for the AI era, designed on the assumption that you are under constant attack. - [Data engineering](https://mg-devs.nl/en/services/data-engineering/): Migrations, data models and pipelines where you can prove afterwards that everything is right. - [Kotlin development](https://mg-devs.nl/en/services/kotlin-development/): Readable, safe backends and APIs in Kotlin, by engineers with over ten years on the JVM. - [Java development](https://mg-devs.nl/en/services/java-development/): Build, modernise and AI-enable business-critical Java systems. - [App development](https://mg-devs.nl/en/services/app-development/): iOS and Android apps from one codebase, with an admin portal that lets you manage the content yourself. ## AI security: four measures in every system An AI model makes no hard distinction between instructions and data, and an agent acts on its own. That is why OWASP ranks prompt injection as the number one risk for AI applications. We build in these four measures by default. - **Least privilege.** An agent gets only the permissions for its task. What it may not do, an attacker cannot make it do either. - **A human approves risky actions.** Merging, deploying and deleting only happen after a human approves. - **External content is data.** Web pages, emails and documents count as untrusted and stay separated from the user's instruction. - **Every action recorded.** An audit log shows who or what made a change, so you can reconstruct afterwards what happened. ## Cases from production - [A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/) (Fintech) - [Shipping software with AI agents, safely and verifiably](https://mg-devs.nl/en/cases/shipping-software-with-ai-agents/) (Software development) - [One platform, seven apps: the carnival platform](https://mg-devs.nl/en/cases/carnival-app-platform/) (Events) - [Carnavalsmaatje: carnival for everyone](https://mg-devs.nl/en/cases/carnavalsmaatje/) (Non-profit) ## Why a small senior team - **You talk to the builders.** The engineer you speak to also designs and builds your system. There is no account manager in between. - **Every change is demonstrably tested.** Each change goes through a pull request, automated checks and a preview environment. Nothing goes live without human approval. - **Security and privacy from day one.** AI processing inside the EU, personal data masked and access only after sign-in. That is fixed in the design before we build. - **Live systems keep working.** We maintain apps whose users never update. That is why we change data and APIs in a backwards-compatible way, always. ## Contact - Email: contact@mg-devs.nl - Web: https://mg-devs.nl/en/contact/ - LinkedIn: https://www.linkedin.com/company/98962780 - Breda, NL · KvK: 91515300 · VAT: NL865679575B01 Source: https://mg-devs.nl/en/ --- # Services We are strongest where AI, data and solid backend engineering meet. Pick the topic that matches your question. ## [AI solutions and agents](https://mg-devs.nl/en/services/ai-solutions/) Assistants on your own data and agents that complete tasks, measurable and under control. We build AI solutions that go beyond a demo: assistants that answer from your own documents and databases, agents that carry out tasks in your processes, and AI features in software you already have. Every solution gets an evaluation, so you know how well it works before you rely on it. ## [Private AI platform](https://mg-devs.nl/en/services/private-ai-platform/) ChatGPT-style convenience for your staff, without company data leaving Europe. A private AI platform gives your organisation its own AI environment: a chat interface for staff, a gateway that decides which model gets which task, and hard guarantees about where data is processed. We design and build that environment, including EU routing, fallback routes, document processing and masking of personal data. ## [AI infrastructure and security](https://mg-devs.nl/en/services/ai-infrastructure-security/) Infrastructure for the AI era, designed on the assumption that you are under constant attack. AI changes the attack surface: models read untrusted text, agents can take actions and automated attacks run day and night. We design and build infrastructure that assumes exactly that: no open ports, least privilege for people and agents, every action recorded, and rules that block when they cannot be guaranteed. ## [Data engineering](https://mg-devs.nl/en/services/data-engineering/) Migrations, data models and pipelines where you can prove afterwards that everything is right. We design and build complex data setups: migrations between databases, data models that keep evolving for years, and pipelines that make data safely available for analytics and AI. What sets us apart is that we build in verification: counts, keys and amounts are automatically compared with the source after every run. ## [Kotlin development](https://mg-devs.nl/en/services/kotlin-development/) Readable, safe backends and APIs in Kotlin, by engineers with over ten years on the JVM. We develop backends, APIs and microservices in Kotlin. Kotlin runs on the JVM, interoperates directly with existing Java code and prevents whole classes of bugs through null safety. We build new systems, migrate Java to Kotlin step by step and add AI features to existing services. ## [Java development](https://mg-devs.nl/en/services/java-development/) Build, modernise and AI-enable business-critical Java systems. We develop and modernise Java systems for organisations where software is business-critical. That ranges from new services to renewing, step by step, an existing system that cannot be allowed to stop. Our experience is in enterprise environments with microservices, integrations and high security requirements. ## [App development](https://mg-devs.nl/en/services/app-development/) iOS and Android apps from one codebase, with an admin portal that lets you manage the content yourself. We build apps for iOS and Android in Flutter, including the admin portal and the backend behind them. Our platform runs several brands from one codebase, each with its own look and its own database. The apps we built have 35,000 active users between them. Source: https://mg-devs.nl/en/services/ --- # AI solutions and AI agents that take over real work > Assistants on your own data and agents that complete tasks, measurable and under control. We build AI solutions that go beyond a demo: assistants that answer from your own documents and databases, agents that carry out tasks in your processes, and AI features in software you already have. Every solution gets an evaluation, so you know how well it works before you rely on it. 73% of Dutch companies that considered AI but did not start name lack of experience as the main reason. One in six companies uses AI in 2025, twice as many as two years earlier. (Source: [Statistics Netherlands (CBS), ICT use by companies 2025](https://www.cbs.nl/nl-nl/nieuws/2025/50/bedrijven-gebruiken-ai-vaakst-voor-marketing-of-verkoop)) ## When to call us - You have a proof of concept that impressed, but never reaches production. - Staff search documents, tickets or databases for answers every day. - You want to add AI to an existing product without rebuilding it. - You are not sure whether your problem needs AI or just good software. Not for those who only need a chatbot widget without a link to their own data or processes; cheaper off-the-shelf products exist for that. ## What you get - **Assistants on your own data (RAG).** Search and answers across your documents and databases, with source citations and per-user permissions. - **AI agents in your process.** Agents that carry a task from start to finish, with limited permissions, logging and a human approval step where needed. - **Ask your database questions.** Text-to-SQL on a shielded, read-only copy, so analysts can request figures in plain language. - **Evaluation and monitoring.** A test set of real questions, measurable quality per version and insight into cost per task. ## How we approach it 1. **Problem first, model second.** We test whether AI is the right solution and what minimum quality is needed for it to be useful. 2. **Start small with real data.** A working first version on a bounded part of your data, measured against a test set. 3. **Make it production-ready.** Permissions, logging, cost monitoring, fallback between models and protection against prompt injection. ## Technology RAG, pgvector, LiteLLM, OpenRouter (EU), Amazon Bedrock (EU), Open WebUI, MCP, Python, Kotlin, TypeScript ## Proof from our own work - Private AI platform for a European financial services firm, with model tiers per task and text-to-SQL on masked data. ([A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/)) - Our own delivery pipeline: AI agents that prepare changes, test them and offer them as a pull request with a preview. ([Shipping software with AI agents, safely and verifiably](https://mg-devs.nl/en/cases/shipping-software-with-ai-agents/)) - A chat assistant in the carnival apps that answers from the organisation's own content. ([One platform, seven apps: the carnival platform](https://mg-devs.nl/en/cases/carnival-app-platform/)) ## Frequently asked questions ### What is the difference between an AI assistant and an AI agent? An assistant answers questions, usually based on your own data. An agent also performs actions: looking up data, creating something in a system, preparing a change. That is why agents get limited permissions and logging. ### What is RAG? RAG stands for retrieval-augmented generation. The model first retrieves relevant passages from your own documents or database and bases its answer on them. That lets it use current, company-specific information and cite its sources. ### How do I know whether an AI solution works well enough? By measuring. Together we compile a test set of real questions and expected answers and measure every version against it. That shows whether a change improves or degrades quality. Source: https://mg-devs.nl/en/services/ai-solutions/ --- # A private AI platform inside the EU, under your own control > ChatGPT-style convenience for your staff, without company data leaving Europe. A private AI platform gives your organisation its own AI environment: a chat interface for staff, a gateway that decides which model gets which task, and hard guarantees about where data is processed. We design and build that environment, including EU routing, fallback routes, document processing and masking of personal data. 49% of Dutch companies that considered AI but do not use it name privacy as a reason. Another 42% name uncertainty about the legal consequences, such as liability for damage. (Source: [Statistics Netherlands (CBS), ICT use by companies 2025](https://www.cbs.nl/nl-nl/nieuws/2025/50/bedrijven-gebruiken-ai-vaakst-voor-marketing-of-verkoop)) ## When to call us - Staff use public AI tools with company or customer data. - You fall under GDPR, DORA or sector rules and must show where data is processed. - You want to use several models without being tied to one vendor. - You want control over AI cost per team or per task. Not needed if your organisation handles no sensitive data and a standard business AI subscription is enough. ## What you get - **Model gateway with tiers.** One entry point with models per task: fast for interaction, cheap for bulk, heavy for analysis. Every tier has a fallback route. - **EU routing that fails closed.** Requests only travel through European endpoints. If a model has no EU endpoint, the request goes to an EU alternative or is refused. - **Documents and search.** Processing of PDFs and office files, vector search in PostgreSQL and a private search engine of your own for web sources. - **Masking of personal data.** Structured data is replaced deterministically; free text passes PII detection before a model sees it. ## How we approach it 1. **Record the decisions.** For every choice (model, provider, region, retention) we record what we choose and why, so compliance can read along. 2. **Build on pinned versions.** Every component runs on a pinned version. An upgrade is a deliberate, reviewed change. 3. **Shield and deliver.** Users only get in through Cloudflare Zero Trust with single sign-on; the AI services publish no ports to the internet themselves. ## Technology Open WebUI, LiteLLM, OpenRouter EU, Amazon Bedrock EU, PostgreSQL + pgvector, Redis, Presidio, Docling, Apache Tika, SearXNG, Docker Compose, Cloudflare Zero Trust ## Proof from our own work - For a European financial services firm: seven model tiers, each with an EU fallback route, in a stack of nine services on pinned versions. ([A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/)) ## Frequently asked questions ### Does our data really stay inside the EU? In our setup prompts and completions are processed only through European endpoints, and routing refuses requests for which that is not possible. Per provider we record what is and is not covered by that guarantee, for example billing and account data. ### Can we use open-source models? Yes, provided they run on European infrastructure. We distinguish data residency from suitability: a model can be hosted safely and still be unsuitable for, say, credit decisions. ### What happens when an AI provider goes down? Every model tier has a fallback route to a second provider inside the EU. Users see a different model at most; the service stays up. Source: https://mg-devs.nl/en/services/private-ai-platform/ --- # AI infrastructure built to withstand attacks > Infrastructure for the AI era, designed on the assumption that you are under constant attack. AI changes the attack surface: models read untrusted text, agents can take actions and automated attacks run day and night. We design and build infrastructure that assumes exactly that: no open ports, least privilege for people and agents, every action recorded, and rules that block when they cannot be guaranteed. Number 1 Prompt injection tops the OWASP Top 10 for LLM applications (2025 edition, the current one), and 'Agent Goal Hijack' tops the OWASP Top 10 for Agentic Applications for 2026. OWASP calls it unclear whether fool-proof prevention exists, and recommends least privilege, human approval for high-risk actions and segregating external content, among other measures. (Source: [OWASP Gen AI Security Project, LLM01:2025](https://genai.owasp.org/llmrisk/llm01-prompt-injection/)) ## When to call us - You deploy AI agents that have access to systems, code or customer data. - Your AI application processes outside input: emails, documents, web pages. - You want to self-host, but without a server reachable from the internet. - An auditor or customer asks how your AI environment is secured. We do not sell standalone penetration tests or certification audits; we design and build the environment so that you are well prepared for them. ## What you get - **Zero-trust access.** Services are reachable only through a tunnel with identity checks. No ports are open to the internet. - **Least-privilege agents.** An agent gets exactly the permissions for its task. Destructive actions require a human and a separate key. - **Prompt injection defence.** Web pages, documents and repository content count as untrusted input: data, not instructions. - **Audit logging and secrets management.** Every mutating action is attributed and logged. Tokens and keys never appear in code, logs or chat messages. - **Hardened containers.** Pinned images, no privilege escalation, internal networks and authenticated connections between services. - **Application security.** Content Security Policy, rate limiting, secure session cookies and device attestation against abuse of public features. ## How we approach it 1. **Threat model.** We map what there is to gain, who can reach it and which input is untrusted, including what a model or agent reads. 2. **Closed by default.** We design from no access and open only what is needed. Rules that cannot be guaranteed block. 3. **Make it provable.** Security rules get tests, changes go through review and the log shows who did what. ## Technology Cloudflare Zero Trust, Cloudflare Access, Docker, Coolify, Linux, Firebase App Check, Play Integrity, DeviceCheck, CSP, GitHub Actions ## Proof from our own work - AI platform for a financial services firm: access through Zero Trust with single sign-on, pinned containers without extra privileges and our own security audit of the server. ([A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/)) - Our own agent environment: agents cannot push, merge or change production data, and every action is in an audit log. ([Shipping software with AI agents, safely and verifiably](https://mg-devs.nl/en/cases/shipping-software-with-ai-agents/)) - Public voting in apps with 35,000 active users, protected against double voting with device attestation. ([One platform, seven apps: the carnival platform](https://mg-devs.nl/en/cases/carnival-app-platform/)) - This website itself: a strict Content-Security-Policy, no trackers, third-party scripts only once you use the contact form, and TLS 1.2 as a minimum. ## Frequently asked questions ### What is prompt injection? In a prompt injection an attacker hides instructions in text an AI model reads, such as a web page, email or document. The model may follow those instructions as if they came from the user. The defence is architecture: treat untrusted input as data and limit what the model is allowed to do. ### What is Zero Trust, and what exactly does Cloudflare Zero Trust do? Zero Trust means no request is trusted because of where it comes from: identity first, then access. With Cloudflare Zero Trust your server opens an outbound tunnel to Cloudflare, so no ports need to be open to the internet, and Cloudflare Access checks per application who signs in, with single sign-on and a second factor. Bots that scan the internet for vulnerable software therefore never reach your application. It does not replace patching, and it only protects what really sits behind the tunnel: an admin screen that is also reachable directly on the IP address stays open. That is why we always verify it from the outside. ### Why is security different for AI than for ordinary software? Because a model makes no hard distinction between instructions and data, and because agents act on their own. A mistake in permissions or input handling therefore has consequences sooner than in classic software. ### Can AI agents safely access our code and systems? Yes, if the environment is set up for it. In our own setup agents work in an isolated working copy, cannot publish anything without checks and human approval, and are not allowed to change production data. Source: https://mg-devs.nl/en/services/ai-infrastructure-security/ --- # Complex data architecture and migrations that match to the cent > Migrations, data models and pipelines where you can prove afterwards that everything is right. We design and build complex data setups: migrations between databases, data models that keep evolving for years, and pipelines that make data safely available for analytics and AI. What sets us apart is that we build in verification: counts, keys and amounts are automatically compared with the source after every run. ## When to call us - You want to move from SQL Server or another legacy system to PostgreSQL. - You want AI or analysts to work on production data without exposing personal data. - Several systems share a data model and every change threatens to break something. - Your data is spread across systems and nobody trusts the numbers. Not for those who only need dashboards; we build the data layer underneath and gladly work with your BI partner. ## What you get - **Database migrations.** Schema-first migration preserving keys, relations and indexes, repeatable and without touching production. - **Masking that preserves consistency.** The same real value becomes the same fake value everywhere, so joins and analyses stay correct. - **Verification as a gate.** A go/no-go check on row counts, exact equality of money amounts, presence of relations and PII leaks. - **Data models that evolve.** Additive, backwards-compatible changes for systems whose old clients keep existing. - **Data ready for AI.** Read-only roles, translated and curated views and vector search, so a model can answer questions safely. ## How we approach it 1. **Understand the source.** We inventory tables, relations, sensitive fields and the questions the data must be able to answer. 2. **Repeatable pipeline.** Restore into a throwaway environment, load, mask, verify, publish. Every step can be re-run. 3. **Live only after proof.** The verification step must pass before anyone, human or model, gets access. ## Technology PostgreSQL, pgvector, SQL Server, pgloader, Presidio, Python, Firestore, Supabase, Redis, Docker ## Proof from our own work - Masked PostgreSQL copy of six SQL Server databases for a financial services firm, with a repeatable pipeline and automated verification. ([A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/)) - One data contract between a live app and an admin portal, spread over separate databases per organisation, without breaking old app versions. ([One platform, seven apps: the carnival platform](https://mg-devs.nl/en/cases/carnival-app-platform/)) ## Frequently asked questions ### How do you migrate from SQL Server to PostgreSQL without risk to production? By never working on production. We restore a backup into a temporary environment, migrate from there and automatically compare the result with the source. The pipeline is repeatable, so a failed run only costs time. ### How do you let AI work safely with company data? With a separate, read-only copy in which personal data is masked and keys and amounts stay intact. The model gets a role with read access to curated views only. ### What does 'backwards compatible' mean for data? That a change does not break existing users. You add fields instead of renaming or removing them, and keep old fields populated for as long as software exists that reads them. Source: https://mg-devs.nl/en/services/data-engineering/ --- # Kotlin development for backends that last > Readable, safe backends and APIs in Kotlin, by engineers with over ten years on the JVM. We develop backends, APIs and microservices in Kotlin. Kotlin runs on the JVM, interoperates directly with existing Java code and prevents whole classes of bugs through null safety. We build new systems, migrate Java to Kotlin step by step and add AI features to existing services. ## When to call us - You are looking for a senior Kotlin developer or small team for a backend or API. - You want to move a Java codebase to Kotlin gradually, without a big bang. - Your backend needs AI features, such as search over your own data or an agent. - Your team lacks experience with architecture, testing or CI/CD on the JVM. There are two of us. We do not supply a team of ten; we do supply a senior who joins temporarily or a bounded project. ## What you get - **Backends and APIs.** REST and event-driven services with a clear domain layer, automated tests and documentation. - **Java to Kotlin.** Step-by-step migration in which Java and Kotlin run side by side and every step can go live. - **Microservices and integrations.** Integrations with payment, CRM and ERP systems, with attention to error handling and idempotency. - **CI/CD and cloud.** Automated build, test and deployment in containers on AWS, Azure or your own infrastructure. ## How we approach it 1. **Step into what exists.** We read the existing code and the team's conventions before proposing anything. 2. **Small, reviewable steps.** Every change is one pull request with tests, so your team can read along and roll back. 3. **Transfer knowledge.** We record choices and work with your developers, so the knowledge stays in your organisation. ## Technology Kotlin, JVM, Spring Boot, Coroutines, Gradle, PostgreSQL, Docker, AWS, Azure, CI/CD ## Proof from our own work - Over ten years of Java and Kotlin in sectors where software cannot be allowed to fail: logistics and ports, mobility, energy and financial services. - The same discipline in other work: additive changes and automated checks on every system we maintain. ([Shipping software with AI agents, safely and verifiably](https://mg-devs.nl/en/cases/shipping-software-with-ai-agents/)) ## Frequently asked questions ### Why Kotlin instead of Java? Kotlin is more concise, has null safety in the type system and offers coroutines for asynchronous work. It runs on the same JVM and uses the same libraries, so your existing Java investment remains usable. ### Can Kotlin and Java live side by side in one project? Yes. Kotlin and Java call each other without a bridging layer. That lets you migrate per file or module and keep shipping in the meantime. ### Can you join our existing team? Yes. We work in your repository, following your review and release process, and take responsibility for a bounded part or advise on the architecture. Source: https://mg-devs.nl/en/services/kotlin-development/ --- # Java development: new builds, modernisation and maintenance > Build, modernise and AI-enable business-critical Java systems. We develop and modernise Java systems for organisations where software is business-critical. That ranges from new services to renewing, step by step, an existing system that cannot be allowed to stop. Our experience is in enterprise environments with microservices, integrations and high security requirements. ## When to call us - You have a Java system that works, but nobody dares to touch it any more. - You want to split a monolith or move it to the cloud without downtime. - You are looking for senior Java capacity for a bounded project. - You want to add AI features to an existing Java application. There are two of us. We do not supply a team of ten; we do supply a senior who joins temporarily or a bounded project. ## What you get - **New builds.** Services and APIs with a clear architecture, automated tests and a deployment pipeline. - **Legacy modernisation.** Renewing step by step: adding tests, updating dependencies, decoupling parts, while the system keeps running. - **Integrations.** Integrations with payment providers, CRM and ERP systems and internal services. - **AI in existing systems.** Search over your own data, summarisation or an agent, connected to your existing Java backend and permission model. ## How we approach it 1. **A safety net first.** Before changing legacy code, we capture the existing behaviour in tests. 2. **Renew in steps.** Every step yields a working system that can go live. No months-long side branch. 3. **Measure and operate.** Monitoring and automated deployment, so you see what happens and can roll back quickly. ## Technology Java, JVM, Spring Boot, Maven, Gradle, PostgreSQL, Docker, Jenkins, AWS, Azure ## Proof from our own work - Over ten years of experience in software development and DevOps, with microservices and enterprise-level security. - For a financial services firm we made existing production data safely available to AI. ([A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/)) ## Frequently asked questions ### Can you take over an existing Java system? Yes. We start with a technical review: code, dependencies, tests, deployment and risks. Then we propose a plan in which the system keeps running while it is improved. ### Do we have to move from Java to Kotlin? No. Modern Java is a fine choice. Kotlin pays off mainly for new builds or when your team wants less boilerplate and more safety in the type system. Both can coexist in one project. ### How do you add AI to a Java application? Usually as a separate service behind your existing permission model. The Java backend remains the source of truth; the AI layer searches, summarises or proposes, and only writes back through existing, controlled APIs. Source: https://mg-devs.nl/en/services/java-development/ --- # App development with Flutter: one codebase, several brands > iOS and Android apps from one codebase, with an admin portal that lets you manage the content yourself. We build apps for iOS and Android in Flutter, including the admin portal and the backend behind them. Our platform runs several brands from one codebase, each with its own look and its own database. The apps we built have 35,000 active users between them. ## When to call us - You want an app for iOS and Android without paying for two teams. - You organise events and want to inform visitors live. - You want to manage your app's content yourself, without a developer. - You want to serve several brands or locations with the same app. Not for games or apps that lean entirely on native 3D or AR; a specialised studio is the better choice there. ## What you get - **App for iOS and Android.** One Flutter codebase, released to the App Store and Google Play, with crash monitoring. - **Admin portal.** Manage news, events, pages, maps and notifications, with roles and permissions per user. - **Interactive features.** Maps with routes and live positions, public voting, a pub crawl with QR codes, a photo booth and push notifications. - **Several brands.** Each brand has its own look, database and configuration, without copying the code. ## How we approach it 1. **Design with real content.** We design with your own content and test with real users. 2. **Build and release early.** Test builds through TestFlight and Google Play from the first weeks. 3. **Maintain without breaking.** Users do not always update. Every data change therefore keeps working for old versions. ## Technology Flutter, Dart, Firebase, Firestore, Cloud Functions, Nuxt, Mapbox, Sentry, TestFlight, Google Play ## Proof from our own work - The carnival platform: seven apps from one codebase and six admin portals, with 35,000 active users. ([One platform, seven apps: the carnival platform](https://mg-devs.nl/en/cases/carnival-app-platform/)) - Carnavalsmaatje: a web application that pairs visitors who need support with a buddy. ([Carnavalsmaatje: carnival for everyone](https://mg-devs.nl/en/cases/carnavalsmaatje/)) ## Frequently asked questions ### Why Flutter? With Flutter you build an app for iOS and Android from one codebase that feels native. That saves development and maintenance cost, and new features reach both platforms at the same time. ### Can I change the app's content myself? Yes. The app comes with an admin portal in which you manage news, events, pages, maps and push notifications, without a developer. ### Can you take over an existing app? Yes, after a technical review of the code and the release pipeline. We ourselves maintain apps that have been in production for years. Source: https://mg-devs.nl/en/services/app-development/ --- # Cases Four projects that show how we work: what the problem was, what we built and what it delivered. ## [A private AI platform for a financial services firm](https://mg-devs.nl/en/cases/private-ai-platform-fintech/) Fintech · 2026 – present A financial services firm wanted its staff to work with AI, without customer data leaving Europe and without depending on a single provider. We built its own AI platform with a model gateway, EU routing that blocks when it cannot be guaranteed, and a masked copy of production data that AI can answer questions about. ## [Shipping software with AI agents, safely and verifiably](https://mg-devs.nl/en/cases/shipping-software-with-ai-agents/) Software development · 2026 – present We built our own environment in which AI agents prepare software changes across seven repositories. An agent creates an isolated working copy, makes the change, runs the checks and offers the result as a pull request with a live preview. Merging and deploying only happen after a human approves. ## [One platform, seven apps: the carnival platform](https://mg-devs.nl/en/cases/carnival-app-platform/) Events · 2023 – present For carnival organisations in the Dutch province of Brabant we built an app platform: one Flutter codebase released as a separate app per organisation, with an admin portal in which the organisation manages news, events, maps and notifications itself. Together the apps have 35,000 active users and simply have to work during carnival, under peak load. ## [Carnavalsmaatje: carnival for everyone](https://mg-devs.nl/en/cases/carnavalsmaatje/) Non-profit · 2023 – present Carnavalsmaatje makes sure that people who cannot go to carnival on their own, such as elderly people and people with a disability, can still take part with a volunteer buddy. We built the web application in which participants and buddies register, are matched and receive their tickets. Source: https://mg-devs.nl/en/cases/ --- # A private AI platform for a financial services firm - **Client:** European financial services firm (name confidential) - **Sector:** Fintech - **Period:** 2026 – present - **Model tiers:** 7, each with an EU fallback - **Services in the stack:** 9, all on pinned versions - **Source databases:** 6 SQL Server databases - **Processing:** European endpoints only A financial services firm wanted its staff to work with AI, without customer data leaving Europe and without depending on a single provider. We built its own AI platform with a model gateway, EU routing that blocks when it cannot be guaranteed, and a masked copy of production data that AI can answer questions about. ## The challenge - Staff needed AI support, but public tools are not an option for financial and personal data. - The organisation had to be able to show where prompts and completions are processed. - Analysts wanted to ask questions about production data, which is full of personal data and lives in SQL Server. ## What we built - **Model gateway with tiers per task.** One gateway with seven model tiers: from fast and cheap for everyday questions to heavy models for deep analysis. Every tier has a fallback route to a second European provider. - **EU routing that fails closed.** All requests are pinned to European endpoints. If a model has no EU endpoint, the EU fallback route takes over; if there is none, the request is refused. Open-source models only run on European infrastructure and are excluded from credit decisions. - **Masked analytics copy.** A repeatable pipeline turns backups of six SQL Server databases into PostgreSQL: restore into a throwaway environment, load, mask, verify and publish. Personal data is replaced deterministically; keys and amounts stay intact. - **Verification as a hard gate.** Before publishing, the pipeline checks row counts, exact equality of money amounts between source and target, the presence of relations and possible PII leaks. - **Search quality measured, not assumed.** An evaluation script measures recall and MRR of search in the knowledge bases. On that basis we chose a deterministic pre-processing step that gives every passage its document title and headings, instead of a costlier variant in which a model rewrites each passage. It needs no model calls and no text leaves the environment. - **A separate, locked-down analyst environment.** Questions to the database go through a dedicated terminal with a role that may only read the masked copy. We added that separation after seeing that a shared terminal could reveal secrets of other services. - **Hardened infrastructure.** Users get in through Cloudflare Zero Trust with single sign-on. The nine services publish no ports to the internet themselves, all images are pinned and containers cannot gain extra privileges. An audit of the running server recorded what was still open and the order in which it can be closed safely. ## Outcome - Staff work with AI in their own environment, without prompts or completions being processed outside Europe. - Analysts can ask questions in plain language about a read-only copy that contains no real personal data. - Every architecture choice is recorded with its reason in a decision document compliance can read. - The organisation is not tied to one AI provider: switching a model is a configuration change. ## Technology Open WebUI, LiteLLM, OpenRouter EU, Amazon Bedrock EU, PostgreSQL + pgvector, SQL Server, pgloader, Presidio, Docling, Redis, Cloudflare Zero Trust, Docker Compose ## Services involved - [Private AI platform](https://mg-devs.nl/en/services/private-ai-platform/) - [AI infrastructure and security](https://mg-devs.nl/en/services/ai-infrastructure-security/) - [Data engineering](https://mg-devs.nl/en/services/data-engineering/) - [AI solutions and agents](https://mg-devs.nl/en/services/ai-solutions/) Source: https://mg-devs.nl/en/cases/private-ai-platform-fintech/ --- # Shipping software with AI agents, safely and verifiably - **Client:** Mardi Gras Developers (our own delivery pipeline) - **Sector:** Software development - **Period:** 2026 – present - **Repositories:** 7 - **Approval:** Always by a human - **Preview:** One environment per pull request - **Deployment:** Only via the pipeline, after merge We built our own environment in which AI agents prepare software changes across seven repositories. An agent creates an isolated working copy, makes the change, runs the checks and offers the result as a pull request with a live preview. Merging and deploying only happen after a human approves. ## The challenge - AI agents can do a lot of work, but an agent with too many permissions is a risk to code, data and production. - Agents read web pages and files that may contain an attacker's instructions. - Clients without a technical background wanted to request changes to their website themselves. ## What we built - **One controlled route to production.** An agent works in its own working copy and cannot push, merge or deploy by itself. One command runs the checks, creates the commit and opens the pull request. - **The right model per task.** The heaviest model does analysis and design, a second model does a critical review, and small mechanical changes go to a cheap model. Every result is checked against the repository's tests. - **A preview per change.** Every pull request gets its own environment on its own web address, so you see the change working before it goes live. - **Security in the environment itself.** Agents are not allowed to change production data and do not deploy themselves. Destructive actions require a human and a separate key. Web content and files count as untrusted input, and every mutating action is in an audit log. - **Changing by conversation.** A client describes a website change in a chat. The environment turns it into a pull request with a preview and sends the link back. A human approves. ## Outcome - Changes arrive as small, checked pull requests with a working preview. - Agents cannot change production data or publish anything without human approval. - Clients request changes in plain language and review the result in their browser. - We set up the same arrangement for clients who want to use AI agents safely in their own development process. ## Technology AI agents, MCP, Git worktrees, GitHub, Coolify, Cloudflare Pages, Docker, Playwright, Sentry, Linux ## Services involved - [AI solutions and agents](https://mg-devs.nl/en/services/ai-solutions/) - [AI infrastructure and security](https://mg-devs.nl/en/services/ai-infrastructure-security/) Source: https://mg-devs.nl/en/cases/shipping-software-with-ai-agents/ --- # One platform, seven apps: the carnival platform - **Client:** Carnival organisations, including Kielegat, Kruikenstad, Oeteldonk, Boemeldonck, Moesland and Carnaval in Uje - **Sector:** Events - **Period:** 2023 – present - **Active users (2026 season):** 35,000 - **Apps from one codebase:** 7 - **Admin portals in production:** 6 - **Screen views in one season:** Almost 2 million For carnival organisations in the Dutch province of Brabant we built an app platform: one Flutter codebase released as a separate app per organisation, with an admin portal in which the organisation manages news, events, maps and notifications itself. Together the apps have 35,000 active users and simply have to work during carnival, under peak load. ## The challenge - Every organisation wants its own app in its own style, but none can afford its own development team. - Usage peaks in a few days a year; nothing may go wrong then. - Users often do not update their app, so old versions stay in use for years. - Public voting on the parade has to be fair: one vote per person. ## What we built - **One codebase, one app per organisation.** The Flutter app is built per organisation with its own theme, configuration and its own Firebase project. New features become available to everyone at once. - **An admin portal generated from a schema.** The portal (Nuxt) generates lists and forms from one data definition. For a new content type you write a schema; the screens follow from it. - **A data contract that does not break.** App and portal only talk through the database. Changes are always additive and backwards compatible, so app versions from years ago keep working. - **Fair voting.** The rule is one vote per device, per app, per event and per category. The server is the only writer of votes; App Check, Play Integrity and DeviceCheck add extra signals against abuse. - **Live float tracking and crowd measurement.** Visitors follow the parade floats live on the map. In Breda we tested a crowd meter: visitors report how busy it is themselves and the app shows that as a grid on the map. The feature can be switched off remotely without a new app version. - **Digital jury scoring.** Jury members score the parade in the admin portal instead of on paper. The result is calculated automatically and can be exported as a report. - **Features for the party itself.** Interactive maps with routes, a pub crawl with QR codes, crowd-density indication with a remote kill switch, a photo booth, a light show and a chat assistant on the organisation's own content. ## Outcome - 35,000 active users across seven apps. - Six organisations manage their own content in their own portal. - One code change delivers a new feature to all apps and portals at once. - The admin portal has more than 1,100 automated tests and the server-side voting functions over 150 more; the database security rules are tested separately. ## Technology Flutter, Dart, Riverpod, Firebase, Firestore, Cloud Functions, Nuxt 4, Vue 3, Zod, Mapbox, Sentry, Docker ## Services involved - [App development](https://mg-devs.nl/en/services/app-development/) - [Data engineering](https://mg-devs.nl/en/services/data-engineering/) - [AI infrastructure and security](https://mg-devs.nl/en/services/ai-infrastructure-security/) Source: https://mg-devs.nl/en/cases/carnival-app-platform/ --- # Carnavalsmaatje: carnival for everyone - **Client:** Carnavalsmaatje - **Sector:** Non-profit - **Period:** 2023 – present - **Online since:** 2023 - **Features:** Matching, registration, tickets - **Hosting:** Cloudflare Pages Carnavalsmaatje makes sure that people who cannot go to carnival on their own, such as elderly people and people with a disability, can still take part with a volunteer buddy. We built the web application in which participants and buddies register, are matched and receive their tickets. ## The challenge - The audience is diverse and not always digitally skilled; registering has to be simple. - Matching participants and buddies cost the organisation a lot of manual work. - It concerns personal data that must be handled with care. ## What we built - **Registration per event.** Participants and buddies register for an event and state their preferences. - **Matching.** The organisation pairs participants and buddies per event in one overview. - **Tickets and agenda.** After matching, both sides receive their tickets and see their agenda. - **Personal account.** Participants and buddies sign in with their own account and manage their profile and registrations themselves. ## Outcome - Participants and buddies arrange registration and tickets themselves, online. - The organisation matches in one overview instead of with loose lists. - The application has been running since 2023 and is maintained by us. ## Technology React, Remix, Supabase, PostgreSQL, Cloudflare Pages, Tailwind CSS, TypeScript ## Services involved - [App development](https://mg-devs.nl/en/services/app-development/) - [Data engineering](https://mg-devs.nl/en/services/data-engineering/) [carnavalsmaatje.nl](https://www.carnavalsmaatje.nl/) Source: https://mg-devs.nl/en/cases/carnavalsmaatje/ --- # How we work A steady rhythm, short lines and something you can look at or use after every step. ## 1. Discovery We map your goal, your users, your existing systems and your data. For AI questions we immediately test whether the problem suits AI, or whether ordinary software is cheaper and more reliable. You get: a scope, the risks and a proposal with planning and price. ## 2. Design We design the architecture and, where there is a screen, the user experience. Decisions about hosting, data residency, models and security are written down with the reason. You get: an architecture sketch, wireframes and a decision log. ## 3. Build We build in short iterations. Every change goes through a pull request, automated checks and a preview environment, so you follow along with working software instead of status reports. You get: working software you can test every iteration. ## 4. Handover and operations After go-live we stay involved: updates, monitoring, security patches and further development. If you prefer to run it yourself, we hand over with documentation and runbooks. You get: documentation, runbooks and a team that knows your system. ## Principles we do not deviate from - **Change additively.** We do not break a live system. New fields and routes are added; existing ones keep working until nobody uses them. - **AI with a human at the wheel.** We use AI agents to build faster. An agent does not merge on its own and does not change production data; a human approves every change. - **Decisions on paper.** Every architecture choice has a recorded reason, so the next team, or you, can later understand why things are the way they are. - **Fail closed.** If a security or privacy rule cannot be guaranteed, the request stops. It does not silently fall through to a less safe route. Source: https://mg-devs.nl/en/approach/ --- # About Mardi Gras Developers A small team of senior engineers building software that keeps working in production. The name comes from Mardi Gras, 'Fat Tuesday'. We started with apps for Dutch carnival associations: software that has to serve tens of thousands of people at once for a few days a year, and simply has to work. Those apps now have 35,000 active users. It taught us to build for peak load, for users who never update and for organisations without an IT department of their own. We now apply the same discipline for companies that want to use AI seriously: private AI platforms inside the EU, data migrations where every amount has to match, and backends in Kotlin and Java. Our background is in business-critical software for logistics and ports, mobility, energy and financial services. ## The team - **Kevin Bos**, Senior Software Craftsman. Specialist in Java and Kotlin, cloud architecture and CI/CD. Has built business-critical systems for over ten years, in logistics and ports, mobility, energy and financial services. Designs our backends, data architecture and AI infrastructure. [LinkedIn](https://www.linkedin.com/in/kevinbos1/) - **Pascal van Hoof**, Full-stack Developer. Builds the apps and web applications, from Flutter to modern web frameworks. Turns complex functionality into something users understand at once. [LinkedIn](https://www.linkedin.com/in/pascal-van-hoof-4b751595/) ## Company details - **Trade name:** Mardi Gras Developers - **Chamber of Commerce (KvK):** 91515300 - **VAT number:** NL865679575B01 - **Based in:** Breda, the Netherlands - **Working area:** The Netherlands and the EU, remote and on site - **Languages:** Dutch and English Source: https://mg-devs.nl/en/about/ --- # AI security check Ten questions about how you use AI. You immediately see a score per area and the three measures that pay off most. It takes about three minutes. Your answers stay in your browser and are not stored or sent anywhere. This is a first estimate, not an audit. The ten questions follow the two current OWASP lists for AI security. Each measure states which risk it addresses. OWASP Top 10 for LLM Applications (2025): https://genai.owasp.org/llm-top-10/ · OWASP Top 10 for Agentic Applications for 2026: https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/ ## 1. Do you know in which region your prompts and completions are processed? - Yes, contractually agreed and technically enforced - Yes, based on what the vendor says - No, or I do not know **Enforce the processing region technically.** Route all model requests through a gateway that only allows European endpoints and refuses requests for which that is not possible. ## 2. Is personal data masked before it goes to a model? - Yes, automatically - Sometimes, manually by staff - No, or I do not know **Mask personal data automatically.** Replace structured data deterministically and run free text through PII detection, so it does not depend on staff paying attention. ## 3. Do staff use public AI tools with company or customer data? - No, there is an AI environment of our own that they use instead - There is a policy, but no control - Yes, or I do not know **Give staff an AI environment of their own.** A ban only works with a good alternative. Offer your own chat environment that is just as easy and where you decide where data goes. ## 4. How is your AI environment or AI application reached? - Only after signing in with single sign-on - With its own password, reachable from the internet - I do not know **Put identity checks in front of the application.** Let nobody reach the application without first signing in through your identity provider, and also check that admin screens are not directly on the internet. ## 5. How does your AI handle outside content such as emails, documents and web pages? - As untrusted and separated from the instruction, or we process no external content - The same as our own instructions - I do not know **Treat external content as data.** Mark everything that comes from outside as untrusted and keep it separate from the instruction, so hidden instructions in a document are not read as a command. ## 6. Do you test your AI application for prompt injection? - Yes, with every release - Once, at delivery - No **Make attack tests part of every release.** Record a set of known attacks and run it automatically on every change, just like ordinary tests. ## 7. What permissions does an AI agent or AI integration have in your systems? - Only what the task needs, or we have no agents - The same as the user or a broad service account - I do not know **Give agents least privilege.** Give every agent its own account with only the permissions for its task. What an agent may not do, an attacker cannot make it do either. ## 8. Is human approval required for risky actions such as deleting, paying or deploying? - Yes, always - For some actions - No **Have a human approve risky actions.** Decide which actions are irreversible or costly and let the AI only prepare them. A human carries them out or approves them. ## 9. Can you reconstruct afterwards what an AI application or agent did? - Yes, there is an audit log per action - Only through general application logs - No **Record every action in an audit log.** Record per action who or what performed it, with which input and with what result, in a place the agent itself cannot change. ## 10. Do you measure the quality of AI answers on every change? - Yes, with a fixed test set - Occasionally, by hand - No **Measure quality with a fixed test set.** Compile a set of real questions with expected answers and measure every version against it, so you see whether a change improves or degrades. ## Pick an attack and see where it stops - **Scanner hunts a known vulnerability.** As soon as a flaw in widely used software becomes public, bots scan the whole internet within hours for servers that still have it. With Cloudflare Zero Trust the server opens an outbound tunnel itself and no port is open; a request without a valid identity never reaches the application, even when it has not been patched yet. Patching is still needed, but it buys you time. - **Stolen password.** Passwords leak through phishing and reuse. Behind Zero Trust a password alone is not enough: sign-in goes through your identity provider with a second factor, per application, and access can be revoked at once when someone leaves. - **Hidden instruction in a document.** An attacker hides a command in a PDF or email that an employee asks the AI to summarise. Because external content stays separated from the instruction, the text is summarised and not carried out. This layer is not watertight, which is why two more follow. - **Manipulated agent looks for secrets.** Suppose an injection does succeed and the agent is told to look up passwords or keys. The agent only has permissions for its task and cannot reach them. - **Agent wants to deploy or delete something.** An agent, manipulated or simply wrong, prepares an irreversible action. It is only carried out after a human has looked at it and approved it. Source: https://mg-devs.nl/en/ai-security-check/ --- # Frequently asked questions Short, direct answers. If your question is not here, just ask. ## What does Mardi Gras Developers do? Mardi Gras Developers is a Dutch software team of senior engineers. We build AI solutions and AI agents, set up private AI platforms and secure AI infrastructure, do data engineering and migrations, and develop backends in Kotlin and Java and apps in Flutter. ## Can you set up AI so that data stays inside the EU? Yes. For a European financial services firm we built a private AI platform in which prompts and completions only travel through European endpoints. Routing fails closed: if a model has no EU endpoint, the request goes to an EU alternative. If there is none either, it is refused. ## Which AI models do you work with? We are model-agnostic. We put a gateway in front of the models, so you choose the right model per task (fast and cheap for bulk, heavy for analysis), with automatic fallback when a provider goes down. That keeps you from being locked in to one vendor. ## How do you secure an AI system? We treat everything a model reads as untrusted input, give agents only the permissions they need and log every action. The AI services publish no ports to the internet themselves, run on pinned software versions and sit in a separate network. Personal data is masked before a model sees it. ## Do you work in existing Java or Kotlin codebases? Yes. We step into existing JVM systems, for example to build new functionality, migrate Java to Kotlin step by step, or add AI features to an existing backend. ## Do you use AI yourselves when developing? Yes, daily. AI agents do analysis, implementation and review for us. Every change then goes through automated checks, gets a preview environment and is only merged after human approval. Agents cannot merge or deploy on their own. ## Who will I be working with? With the engineers themselves. Kevin Bos and Pascal van Hoof do the design, the build and the communication. There is no account manager in between. ## Can I also hire you as a freelancer? Yes. Besides projects with an agreed outcome, we join existing teams as freelance senior engineers, for example for Kotlin and Java backends, data engineering or setting up AI infrastructure. Availability and duration are agreed per assignment. ## What does a project cost? That depends on complexity, functionality and lead time. After an intro call and a short discovery you receive a proposal with scope, planning and price. We help you work out what delivers the most value within your budget. ## How does a collaboration start? With an intro call of about 30 minutes. We discuss your question and tell you honestly whether we are the right party. After that comes a discovery in which we record scope, risks and approach. ## Do you offer maintenance after delivery? Yes. We take care of updates, security patches, monitoring and further development. We have been maintaining the carnival apps and admin portals we built for years. ## Where does the name Mardi Gras Developers come from? From Mardi Gras, 'Fat Tuesday'. We started with apps for carnival associations. The name stayed; the work broadened to AI, data and backend development. Source: https://mg-devs.nl/en/faq/ --- # Book an intro call Tell us briefly what you are dealing with. Kevin or Pascal replies personally. 1. You send a message with your question. 2. We schedule an intro call of about 30 minutes. 3. You receive a proposal with scope, planning and price. ## Contact - Email: contact@mg-devs.nl - Web: https://mg-devs.nl/en/contact/ - LinkedIn: https://www.linkedin.com/company/98962780 - Breda, NL · KvK: 91515300 · VAT: NL865679575B01 Source: https://mg-devs.nl/en/contact/