From First Call to Production
A bespoke engagement is a high-trust commitment. We make the path legible before we ask you to take a single step on it. Four phases, each with a defined duration, a defined deliverable, and a clear way out.
What each phase actually contains.
Frame
1–2 weeks· No cost, no commitmentWe listen. You leave with a written framing.
We start with a single conversation and a written problem framing. We map your operational challenge against the constraints that actually matter — sovereignty, data residency, governance accountability, deployment topology. You leave with a 3–5 page memo describing the problem in our shared vocabulary. It is yours to keep, even if we go no further.
- ▸60–90 minute discovery conversation with your operational and governance leads.
- ▸Written framing of the problem in plain institutional language, free of vendor terminology.
- ▸Initial constraint map: data sovereignty, regulatory exposure, deployment topology, security posture.
- ▸Honest assessment of fit. If we are not the right partner, we tell you, and we tell you who is.
Scope
2–3 weeks· Fixed fee, fully credited if we proceed to BuildWe architect. You receive a fixed-fee proposal.
Architecture sketch, success criteria, deployment model, risk register. We name what we will build, how it will be measured, where it will run, and what we will not do. Procurement gets a single line item; your team gets a document that survives a leadership change.
- ▸System architecture sketch, including data flow, model selection, and human-in-the-loop checkpoints.
- ▸Defined success criteria — measurable, not aspirational.
- ▸Deployment model: cloud, on-premise, or air-gapped, with full cost and operational implications.
- ▸Risk register with mitigations: model failure modes, data exposure, governance edge cases, exit conditions.
- ▸Fixed-fee build proposal. The Scope fee is fully credited against Build if you proceed.
Build
6–16 weeks· Weekly cadence, transparent throughoutWe ship. You see it grow each week.
Lab-tested prototype to hardened deployment, on your infrastructure or ours — your choice. Code, decision records, and weekly working sessions throughout. Nothing about the system is locked away from you. When it goes live, you own the source, the configuration, and the operational knowledge to run it.
- ▸Lab-tested prototype before any production code is written.
- ▸Weekly working sessions with your designated counterparts. No status decks; working software.
- ▸Decision records published as the system is built — you see why each architectural choice was made.
- ▸Full source delivered as we go, on your repository. No black boxes, no proprietary runtime.
- ▸Acceptance testing against the success criteria defined in Scope. We do not declare done; you do.
Hold
Ongoing· Optional, never requiredWe maintain. You can leave at any time.
Maintenance, model updates, governance reviews, localization. Optional, never required. You can in-source operations whenever you are ready — the system is yours, the documentation is complete, and our exit is part of the deliverable from day one.
- ▸Patching, dependency upgrades, and model refreshes as the underlying landscape moves.
- ▸Periodic governance reviews — does the system still match the institution's policy posture?
- ▸Localization and adaptation as the deployment footprint grows across regions or mandates.
- ▸In-sourcing on request: full handover to your team, including documentation and training, at any time.
What this engagement model actually produced.
HRIA · Built for the United Nations Development ProgrammeOver recent months we refined and deployed HRIA.eu with the UNDP team, focused on operationalizing Human Rights Impact Assessments in a more structured and usable way.
We deliberately did not build the system as a fully automated AI assessment engine.
Technically, that was possible — and in some use cases where speed and economies of scale are decisive, it would even be the right answer. But governance processes are not just about outputs. They are also about building institutional judgment and what we call human muscle memory — the ability to understand, reason, and conduct assessments consistently over time.
Instead of replacing the process, HRIA.eu was designed as a structured digital methodology layer — and, in the future, potentially an AI co-pilot supporting human decision-making rather than bypassing it.
Recent feedback from DPA officers and institutional users across Europe and beyond has been encouraging and, more importantly, operationally useful. It drove improvements to:
- ▸stakeholder mappings,
- ▸rights descriptions,
- ▸usability,
- ▸methodological consistency,
- ▸and localization for partner regions.
Building trustworthy governance systems is less about “AI magic” and more about helping institutions retain coherence and operational memory — with humans in the loop.
The conditions, written before the code.
Every term that procurement and legal need is decided up front. None of this is negotiated under deadline pressure once a system is already deployed.
Intellectual property
You own the code, the configurations, the prompts, the fine-tuning data, and the deployment artifacts. We retain no rights and no operational dependencies on your infrastructure.
Data handling
Your data does not leave your environment unless you explicitly authorize it. We design for air-gapped deployment by default and treat cloud as a deliberate choice, not a default.
Exit clause
Termination terms are written into the Scope contract before a single line of production code is shipped. No retention penalties, no transition fees beyond reasonable handover time.
Procurement structure
We can engage as a single-source vendor, under existing framework agreements, or via consortium arrangements with established institutional partners. Tell us your procurement vehicle; we will fit into it.
Phase I is free.
Tell us the operational problem in two sentences. We will reply with a date for the framing call and a short list of what to bring.