Software, data and AI engineering that holds up in production.
A small senior team of data, software and DevOps engineers. We build the systems that turn your data into decisions, and test applied AI in our own lab before it reaches yours.
Research that has to survive contact with production.
Our lab works on retrieval, evaluation and industrial forecasting, and takes consulting missions where that work is the whole point. What it learns goes straight into client projects.
Pipelines, warehouses and reporting your teams can trust, so the same figure means the same thing in every report.
For technical teams
Data contracts between teams, lineage that survives schema changes, governance for GDPR and sector regulators. Warehouse or lakehouse, chosen on evidence.
Independent advice on strategy, architecture and AI, framed in business terms: total cost of ownership, return on investment and risk. From senior engineers who can also build what they recommend.
Advisory work usually starts with one of these situations.
Your systems slow the business down
Every change takes months and the run budget keeps growing. We map what exists, quantify the technical debt and give you a clear modernise, replace or rebuild decision.
You face a large technology investment
A new platform, a cloud move or a major vendor contract. We build the business case for each option: TCO, CAPEX and OPEX profile, payback period and exit costs.
Leadership wants an AI plan
We rank use cases by expected value and feasibility, check your data is ready, and propose a phased investment with clear go/no-go gates.
A project is off track
Budget overrun, deadline missed, confidence eroding. We run an independent review and tell you plainly what to stop, fix and keep, with the cost of each option.
What we advise on
Technology strategy and roadmap
A multi-year plan tied to business objectives, with priorities, a budget envelope and milestones your steering committee can track.
Architecture review
An assessment of your systems, integrations and technical debt, with the risk and cost of leaving each issue unresolved.
Build, buy and vendor selection
Independent comparison of products and partners on TCO, lock-in and delivery risk, through to contract review.
AI strategy
Use cases ranked by value at stake and feasibility, with a governance model, an investment profile and a first pilot.
Cloud and data platform choices
Which platform, which migration path, and how to keep run costs predictable with clear FinOps rules.
Technical due diligence
For acquisitions, investments or partnerships: what the technology is really worth, and the investment it will need next.
Three ways to work with us
2 to 6 weeks
Assessment
A fixed-price study ending with a written recommendation, a business case and a costed roadmap.
A few days a month
Ongoing advisory
A senior engineer at your side for architecture, vendor and budget decisions, without the cost of a full-time hire.
From recommendation to production
Advice and delivery
The same team that wrote the plan builds it, so nothing is lost in the handover.
Custom software, built for how your business actually works.
When off-the-shelf tools no longer fit, we design, build and run the applications your teams depend on, and connect them to the systems you already have.
Most companies do not lack data. They lack data they can use: scattered across systems, defined differently in every report, and too unreliable to build AI on.
A model in a notebook is not a product. We take your use case from feasibility study to a monitored model running in your operations, in the cloud, on site or on the device.
Generative AI, connected to your systems and governed from day one.
We take AI beyond the pilot: assistants and agents that work with your documents, your ERP and your CRM, with the security and audit controls your risk and compliance teams will ask for.
Answers from your procedures, contracts and documentation, with the source attached.
Document processing
Invoices, claims, fines and contracts read, checked and routed automatically.
Customer and employee support
Assistants that resolve routine requests and hand complex ones to a person.
Agents in your business systems
Agents that create orders, update records or prepare reports in your ERP and CRM, under supervision.
What happens behind one request
Prompt from a user or an agent
A question in chat, a task for an agent, or a call from one of your applications.
AI gateway and guardrails
Checks identity and permissions, filters prompt injection and sensitive data, applies token budgets and logs every call.
Retrieval (RAG)
Hybrid vector and keyword search over your documents, reranked, and limited to what this user is allowed to see.
LLM orchestration
The right model for each task (Azure OpenAI, Mistral or an open-weight model on your own infrastructure), with tool calling, structured outputs and automatic failover.
Grounded answer or agent action
Returned with citations, checked against its sources for hallucinations, and scored by an evaluation pipeline over time.
Governance, built in
Prompts, embeddings and logs stay in your own cloud tenant.
Retrieval respects the permissions you already have.
Every prompt and model call is traceable to a person or a service.
Designed with GDPR and the EU AI Act in mind.
Token costs tracked per team and per use case.
-39%
claims processing time after the first agents went live
No catalogue. Tell us the topic and the people in the room, and we design the session around them.
Sessions we run every month.
For companies, public bodies and training partners, from a half-day briefing to a multi-day hands-on course.
English, French, Dutch
Taught in the language of the room.
On-site or live online
Across Belgium and Europe.
In-house or through partners
For one team, or open sessions run with training organisations.
Recently delivered
Client names on request
{{ trMarquee }}
And much more.
Generative AIPromptingCopilot for Microsoft 365Power BIDAXExcel automationPythonpandasSQLdbtSparkAirflowDatabricksSnowflakeAzureAWSDockerKubernetesTerraformGitLab CIMLOpsMLflowLLM applicationsRAGAgentsData governanceData qualityData modellingGDPR for data teamsEU AI ActAPIsTypeScriptReactJavaSpring BootTestingObservabilityFinOpsCybersecurity basicsAgile deliveryComputer visionForecasting
A small AI lab with a customer waiting at the end of every experiment.
A few FP2K engineers who spend part of their time on applied AI research, and take on consulting missions for organisations whose hardest question is not which model to use but whether the answer can be trusted.
Our opinion on the techniques, tools and platforms we meet in projects, updated twice a year. Each entry is one line of judgement, not a sales pitch.
{{ rg.name }}
{{ rg.hint }}
{{ it.name }}{{ it.cat }}
{{ it.note }}
Free first step
Bring us your question. We come back with hypotheses.
Before any proposal, we study your case and bring two or three concrete hypotheses to a 30-minute call. You keep the thinking, whether or not we work together.
A few lines or a short call: the decision you face, the data you have, the constraints.
2
We do the homework
We look at similar problems, the approaches that work and the traps we already know.
3
30 minutes, 2–3 hypotheses
Where we think the answer lies, how to test it, and what it would take. Yours to keep.
Research areas
Retrieval
Grounded retrieval & citation integrity
How a system proves where an answer came from, and refuses to answer when it can't.
Evaluation
Evaluation under distribution shift
Benchmarks decay the moment a system meets real traffic. We build evaluation that keeps telling the truth eighteen months after launch.
Forecasting
Industrial time-series forecasting
Sparse failure events, irregular sampling and sensors that lie. Applied to industrial and operational data.
Efficiency
Small-model inference economics
When a 3B model on your own hardware beats a frontier API on cost, latency and data residency — and the honest cases where it doesn't.
Consulting missions
The Lab takes on a few missions a year, working directly with the client's own team. These are not staffing arrangements — they are questions we agree to answer.
Every figure below was measured against a baseline agreed with the client before we started. Named references available on request.
{{ c.capability }}{{ c.sector }}
{{ c.title }}
{{ c.body }}
{{ c.m1 }}
{{ c.l1 }}
{{ c.m2 }}
{{ c.l2 }}
About
We started with an argument about handovers.
Since 2014, we have been a small team of engineers who watched too many good systems die the moment they were handed to someone else. At FP2K, the people who design a system stay on it.
Scope
Build
Run
The usual way
ConsultantsDevelopersSupport desk
The FP2K way
Same engineers, start to finishYour team takes over
Seniority is not optional
The median FP2K engineer has eleven years of experience. We do not sell junior time at senior rates, and we do not pad teams.
ADR-014 · Event store
Chosen Postgres + outbox
Rejected Kafka, team too small
Rejected Managed queue, lock-in
Evidence over opinion
Architecture decisions are written down with the alternatives we rejected and why. If we were wrong, the record says so.
FP2K hours per month
Leave, don't linger
Success is your team not needing us. Two of our longest engagements ended because the client's own team outgrew the need.
Who you work with
No account layer, no org chart to climb. The people who scope the work are the people who do it.
{{ ro.title }}
{{ ro.text }}
{{ tl }}
Careers
Stay on the system long enough to be proud of it.
We hire slowly and keep people for a long time. In a small team, your work is visible and your opinion counts.
Time for the Lab
Every engineer gets one day per fortnight inside the Lab, on Lab problems.
No billable-hour targets
Utilisation is a planning input, not a performance metric. Nobody is measured on it.
Learn and speak
Training and conference time for everyone, not only the Lab.
Open salary bands
Published internally, reviewed annually against market data we share with you.