Case Studies

RACE Programming in commercial delivery

Real engagements run in the shape of RACE Programming: what was delivered, how fast, and what the model bills for. Client identities are withheld; the delivery facts are not.

Commercial delivery

AI-augmented supply-chain product

A fixed-price, end-to-end delivery for a US-based supply-chain client. A working prototype produced between sales calls, a foundation-level redesign absorbed mid-engagement, and four Definition-of-Done gates green every Stint.

3-person team · ~3 months PoC → production React · Python · Ridge-regression ML Paid follow-on engagement
Read the case study →
Capability demonstration

Two native apps to one React Native codebase

A three-person team rebuilt two separate native mobile apps as a single React Native app from nothing but the old source archives, producing a working, near-pixel-perfect demo in three working days, then one-day Stints.

3-person team · 3 working days to first demo React Native · Expo · AI-generated tests One-day Stints
Read the case study →
AI product delivery

An embedded AI knowledge assistant

A two-person AI team shipped a production, document-grounded AI assistant for a 50+ office professional-services network, at an estimated ~3× the speed and ~⅓ the cost of a conventionally-staffed team.

2-person AI team · ~15 vs ~43 working days (est.) Claude · grounded RAG · zero data retention Live across 50+ offices
Read the case study →
AI-assisted data operations

Agent-to-agent data imports

A multi-office US property valuation firm consolidates data from every regional office without mandating a template. When the sender is an AI agent, the import service answers in machine-actionable findings, including a verdict on whose problem it is.

Pit Wall of two, no Pit Crew ~10× work per budget · ~5× faster (est.) In production
Read the case study →
Autonomous software maintenance

A one-person orchestra, and the merges nobody does anymore

One engineer automated the upstream merge work that a fork with two dozen feature branches would otherwise demand from a team. Scripts plan, the agent resolves in a sandbox, and a human arbitrates only what genuinely needs it.

One engineer + the agent ~7 of 8 merges untouched Runs on a schedule
Read the case study →
Agent infrastructure

Agents that cannot leak credentials

Container isolation protects the host, not the secret inside it. This platform moves the credential boundary to the network, so the agent reaches every system it needs and still never holds a real token.

Substitute tokens, swapped at the edge SSH isolated, not substituted Detection without damage
Read the case study →