Architecture review
We review an existing AI system or design: its evidence path, saved state and failure modes.
We also check authority boundaries. You receive a written risk report and a practical plan.
typical: 2–3 weeksMenu
WORK WITH US / ENGAGEMENTS
We build governed AI systems for other people's decisions.
For organizations where AI answers need sources and actions need owners.
Describe your decision problemFOUR WAYS TO ENGAGE
Choose the smallest useful commitment.
Start with the smallest format that can produce a decision.
System build
We design and build from the decision problem to a working governed loop.
You receive an inspectable MVP, documentation and a complete handoff.
typical: from 6 weeks to MVPApplied R&D pilot
We run a pilot in your domain and data, with success criteria agreed first.
Either side can stop when results do not support continuing.
typical: 4–8 weeksSupported operation
After launch, we watch system behavior, handle operational failures and improve answer quality and source support.
Planned support only; it is not round-the-clock on-call coverage.
typical: monthly, ongoingWHO THIS IS FOR
Teams whose decisions need a visible trail.
These situations need explicit system controls.
- 01
Thousands of documents, but every answer must point to a specific source.
- 02
A counterparty must be checked and the final decision kept with its supporting record.
- 03
Decisions happen in messages, and half the agreements lose their owner.
- 04
Research runs for months, and every conclusion needs a visible origin.
HOW AN ENGAGEMENT RUNS
Decision gates are part of the work.
At every stage, you can see what has changed and decide whether to continue.
- 01Frame
We define the decision, users, sources and cost of failure; you receive a brief and choose whether to continue.
- 02Contract
You receive the scope, success criteria and stop conditions before build work.
- 03Compose
We assemble the smallest useful system; you receive the first working loop.
- 04Challenge
We test failures together; you receive results and decide whether to stop or revise.
- 05Operate
You receive the system, documentation and operating plan, then decide whether to expand.
WHAT YOU GET
A system your team can inspect and keep.
- 01
An inspectable system, not a black box.
- 02
An evidence path for every material answer.
- 03
Replayable state and documentation for recovery.
- 04
A handoff without vendor lock; self-hosted by default.
ENGAGEMENT BOUNDARY
The lab is in private mode and takes a small number of engagements.
We don't sell model access or generic chatbots, and nothing here is investment or legal advice.
COMMON QUESTIONS / PUBLIC ANSWERS
What to know before the first conversation.
Short answers about the lab, engagement boundaries and private mode.
- 01What is ROOM 302?
- ROOM 302 is an applied AI systems lab that builds inspectable systems for consequential work. It connects sources, saved state, specialist roles and human approval.
- 02What kinds of systems does ROOM 302 build?
- ROOM 302 builds research, knowledge, operations and decision-support systems. The public registry shows current programs and their maturity.
- 03Is ROOM 302 a chatbot or API provider?
- No. ROOM 302 designs complete working systems; it does not sell generic chatbots, model access or a public API.
- 04How does an engagement start, and how long does it take?
- Work can start as a 2–3 week architecture review or a 4–8 week pilot. A system build typically starts at six weeks to an MVP.
- 05What happens after launch?
- Supported operation is available as planned monthly work. It covers system behavior, operational failures, source support and documented improvements, not round-the-clock on-call coverage.
- 06How are project data and residency handled?
- Projects are self-hosted by default, with data handling and residency agreed before work begins. Private source material does not enter the public site.
- 07What does ROOM 302 not provide?
- ROOM 302 does not provide investment or legal advice, sell model access, or publish private dossiers. Human owners retain consequential decisions.
- 08Can someone approach ROOM 302 anonymously while the lab is in private mode?
- The lab is in private mode and the public site omits personal identities and direct contacts. Anonymous intake remains closed until first-party storage is configured.
START WITH THE DECISION