My projects

110-agent finance platform where a deterministic policy engine runs tenant, capability, and approval checks before any LLM call. $0 default autonomous spend, database-enforced immutable audit log.

View siteRepo

Trust contracts as JSON Schema: risk tiers, tool permissions, deny-by-default egress, approval gates, compliance mappings. Prompt-injection threat model with adversarial fixtures, ~28 CI gates.

View siteRepo

Multi-agent debugger with zero LLM calls in the analysis path: causal graphs, weighted root-cause scoring, failure prediction. 206 zero-mock tests, zero runtime deps, no network egress.

View siteRepo

357-agent orchestrator governed by per-agent budget caps, a pattern-matched bash security gate, auto-downgrade daily budgets, a hard per-run spend ceiling, and correlation-ID telemetry.

View siteRepo

Time-travel debugging for agent traces: zero-trust import parsing with signature verification and size clamps, five trace formats auto-detected, and a divergence/latency/token diff engine.

Deep-research agent with a real human-in-the-loop approval workflow, where the run suspends until a human approves, plus a deterministic citation audit with zero AI in the verification path.

Repo

Governed agent runtime: kill switch with cost and error auto-triggers, circuit breakers, a four-tier execution sandbox, hard daily and per-task budget caps, and a dead-man's switch.

Company-scale agent platform on a three-plane architecture covering control, data, and evidence. State-machine status transitions, hard execution timeouts, adversarial eval harness with a 33-task golden suite.

Repo