Freelance AI Engineer & Software Architect · Netherlands
I design and build production AI systems — from retrieval and intelligent workflows to the dependable Python software that makes them useful, measurable and maintainable.
Some projects need an independent technical perspective. Others need someone who can make architectural decisions and write the code. I’m comfortable doing both.
Assess opportunities and existing systems, challenge assumptions, choose suitable architectures and design realistic paths to production. Useful before a large investment or when a team needs a second pair of experienced eyes.
Feasibility · Architecture reviews · AI strategy · Evaluation designWork alongside your developers or take technical ownership of a workstream: architecture, implementation, code reviews, infrastructure and the operational details that make the system reliable.
Fractional technical leadership · Python · AI systems · DeliveryTurn unclear requirements into working software, validate ideas with focused prototypes, or make an existing platform easier to extend. The aim is to learn quickly without leaving behind an unmaintainable prototype.
Discovery · MVPs · Workflow digitisation · Platform evolutionA selection of work across AI engineering, retrieval, evaluation and long-lived platforms. Each example reflects work I contributed to or led, not necessarily a solo build.
For the Dutch business register
A conversational assistant supporting tens of thousands of business registrations annually. I helped move its core from Rasa to an Azure OpenAI architecture, using stateful orchestration and validated structured outputs, and owned the CI/CD pipeline into Kubernetes.
Python, Azure OpenAI, Kubernetes, Terraform, GitLab CI
For the Dutch business register
Built a classifier combining semantic and keyword search to map free-text company descriptions to a national activity taxonomy. Delivered through an internal API for reuse across teams.
Python, Postgres with pgvector and ParadeDB, Azure
For a Django agency
Built a hybrid-indexed knowledge system spanning engineering and communication tools. Answers include source references and are available in web, Slack, Jira and developer workflows via MCP.
Django, Postgres with pgvector, MCP, Slack and Jira integrations
Shareholder & technical lead · preventive health platform
As a shareholder and technical lead, I help shape the product’s technical direction and lead hands-on development. The platform drafts personalised plans from approximately 90 health metrics, with deterministic reference checks and mandatory clinician approval. Beyond AI, it supports consultations, payments and patient-record integration.
Django, Wagtail, Postgres, Celery, Docker
For a university medical centre’s image analysis group
Spent four years contributing to grand-challenge.org, which evaluates containerised medical imaging algorithms on withheld test sets. Worked across the platform, AWS deployment pipelines and clinician reader studies.
Django, Celery, Docker, AWS (S3, Lambda, ECS and Batch)
Issues move past three stations, refine, develop and review, each run by an agent with its own identity. The refiner checks an issue against the definition of ready and asks its author one question at a time; the reviewer sends work back with requested changes; anything stuck goes to a person. People write issues, answer questions and ship, all visible on a live dashboard.
Python, Django, coding agents, GitLab and JiraGiven the brief, the functional, UX and technical designs and a starting repository, the system plans a project into work items, then implements, reviews and verifies each one with coding agents, and compares the result with the project as it was really built. Each task gets only the capabilities it needs over MCP, and every action leaves an audit event.
Python, Django, Docker, MCP, CodexA shared core that encodes who may do what and why, for people, agents and service accounts alike, and keeps an account of what happened. An existing platform moves onto it one process at a time while it keeps running. In the same spirit, a customer portal where access follows an explicit written binding instead of a tracker’s permission scheme.
Python, Django, Jira and GitLab adaptersAround five hundred users get training programmes and answers grounded in a hybrid index, dense and sparse retrieval fused by reciprocal rank, and a form check from pose estimation on their phone’s own camera. Architecture and build shared with two fellow founders; the retrieval stack is mine. The social pipeline that promotes it grew into Viral Foundry.
Python, React Native, MediaPipe, hybrid retrievalDescribe an idea once and get the whole video: script, consecutive scenes, and the same actors and voices across videos, published to social platforms. A pipeline that orchestrates text, speech, image and video models, with a queue that resumes after a restart and tests that never reach a paid API.
FastAPI, Postgres, Redis, React, generative media APIsFifteen issues a developer solved within four hours and twenty-two production errors, each with the repository at the commit before the fix and the real fix kept hidden as the reference. Agent setups are judged on the work the company actually does, not on public leaderboards.
Python, Git, Jira, GitLab and Sentry dataEditors pick published pages from the page tree and an audit runs in the background, with results that point out which issues Wagtail’s own checks also flag. Access is a permission you grant like any other.
Wagtail, Celery, axeMy own products are where I can explore ideas end to end—from a technical hypothesis to something people can use. They inform how I approach client work.
An AI-powered personal trainer bringing strength training, nutrition, recovery and exercise-form analysis into one coaching experience. I co-founded the product and worked on its architecture, retrieval-grounded coaching and mobile AI capabilities.
It’s a reminder that AI features only matter when the complete product is genuinely useful—and the surrounding software is dependable.
An AI content studio that creates videos, carousels and stills from a brief, publishes to connected social channels and uses engagement data to inform what comes next. Its workflows coordinate multiple generative models and keep assets consistent across scenes. I build it together with a co-founder.
For me, the interesting challenge is building robust, recoverable workflows around powerful but imperfect models—not simply calling an image or video API.
Two complementary strands of work: making day-to-day operations more coherent, and rethinking the foundations on which business software is built.
I’m working with a creative event and catering organisation to formalise and digitise how work moves from client enquiries and event planning through purchasing, kitchen production, staffing and logistics. The work starts by mapping real processes and the information each team needs, then translating them into an integrated, maintainable operational platform.
The objective is not automation for its own sake. It is clearer ownership, consistent data and handovers, less manual coordination and a foundation that can evolve as the business grows. This transformation is ongoing.
I’m exploring a declarative runtime for business applications. Instead of repeatedly rebuilding data models, authorization, audit trails, migrations and workflow infrastructure, an application defines its resources, actions and policies, while the runtime enforces the rules.
A major motivation is AI-assisted development: letting people and coding agents express business intent without granting them unrestricted control over production data or security boundaries. This is a design exploration, not a finished platform.
I can help at the architecture stage, build with your team, or take ownership of the engineering required to get an existing system production-ready.
Connect models to real processes, tools and APIs. Design scoped permissions, clear approval points, structured model interfaces, observability and safe failure modes.
Turn fragmented data into grounded answers or useful classifications, combining sparse and semantic retrieval, ranking, source attribution and evaluation.
Build representative test sets, compare configurations, track regressions and establish where automation performs well enough to earn trust.
Develop and improve Django or FastAPI systems, integrations, background jobs, deployment pipelines and infrastructure designed for ongoing maintenance.
Senior engineering is as much about judgment, communication and knowing what not to build as it is about writing code.
Test against representative cases, inspect failures and be explicit about what has and hasn't been verified.
Design automation around real operational boundaries, with clear permissions and review at consequential steps.
Prefer straightforward architecture, readable changes, repeatable delivery and software the next developer can confidently maintain.
I'm a Netherlands-based freelance AI engineer and Python developer with a background in artificial intelligence and more than twelve years of software engineering experience. My work has ranged from research infrastructure for medical imaging to AI systems used in business registration, preventive health and developer workflows.
Alongside consulting and client work, I build products such as SensAI.PT and Viral Foundry. Those projects let me experiment with emerging capabilities and learn what it takes to turn a technical idea into a usable product. I enjoy the point where applied AI meets solid software engineering: translating an ambiguous problem into a system that is understandable, testable and genuinely useful for the people relying on it.
Tools I commonly use: Python · Django · FastAPI · PostgreSQL · pgvector · ParadeDB · Celery · Redis · Docker · Kubernetes · Azure · GitLab CI · React · MCP
Whether you need an architecture review, help digitising a business process, a technical lead or someone to build a production AI system, tell me what you're working on. Dutch or English, remote or on-site in the Netherlands.
mike@mikeoverkamp.nl ↗