On the bench
right now.
What I'm architecting, writing up and testing next — case studies from recent builds and the agent capabilities moving through the workshop.
Now
November 2026- Writing up|Case study
Stratton Oakmont — case study
The market-intelligence build, end to end: how filings and market data are ingested, scored and turned into a daily signal, and where the model stops trusting the data.
Market intelligence · 12-day build
- Writing up|Case study
Vin Check USA — case study
Orchestrating several vehicle-history providers behind one lookup: caching, partial-failure handling and turning three inconsistent response shapes into one report.
Automotive data · API orchestration
- In build|Agent capability
Entity-grounded agent retrieval
Agents that answer from a client's own records instead of a stuffed prompt — scoped server-side queries, permission checks, and citations back to the record they used.
Agent architecture
Next
December — January- In build|Case study
FieldIQ — case study
Dispatch and scheduling for field teams: how job allocation, technician availability and customer windows are reconciled without a scheduler on the phone.
Field service · Scheduling
- In build|Case study
TablePulse — case study
Restaurant operations in one screen — covers, shifts and service reporting — and the data model that keeps front-of-house and kitchen on the same numbers.
Hospitality · Operations
- Exploring|Agent capability
Approval queues for agent writes
Human-in-the-loop before an agent commits anything: draft, review, approve. The agent proposes the record change, a person signs it off, the log keeps both.
Agent guardrails
- Exploring|Agent capability
WhatsApp-channel agents
The same agent, same tools and same guardrails, reachable over WhatsApp — with conversation state and hand-off to a human when confidence drops.
Channels
Later
2027- Exploring|Case study
ClaimAssist and Rebody — case studies
Two teardowns queued: claims handling with document extraction, and the body-composition tracker's measurement model.
Case studies backlog
- Exploring|Agent capability
Scheduled agents with memory
Agents that run nightly, remember what changed since yesterday and report only the delta — instead of a fresh summary nobody reads.
Automation
- Exploring|Agent capability
Multi-agent hand-off patterns
When one agent should hand work to another: shared state, tool boundaries, and how to keep the audit trail intact across the hand-off.
Agent architecture
Want yours on this list?
I take a limited number of builds at a time so each one gets a written-up teardown. Send a brief and I'll tell you where the next slot lands.