Work / Showcase
Work that
leaves a receipt.
Six systems, one discipline. Each card opens its own case page — what it is, what it is built from, and what a reader can check for themselves.
Agent reliability · Open source
2026
EvidenceForge
An open-source framework for building auditable, evidence-citing AI agents.
Outcome: Every claim in an agent run can be traced to the evidence behind it.
Read the case →
API tooling · Developer product
2026
mimicplus
Record a HAR, get a versioned spec, a typed Python client, OpenAPI, and a breaking-change check.
Outcome: An API you recorded becomes a spec you can test, not a screenshot you forget.
Read the case →
Multi-agent systems
2026
Governed Agent Engine
A governed multi-agent engine: Termux-installable, stdlib plus PyYAML, with spec binding and audit.
Outcome: Agents that run on a phone and still answer to a spec.
Read the case →
Audit infrastructure
2026
Verity
An append-only evidence ledger for agent runs. Claims must cite evidence.
Outcome: History that cannot be quietly rewritten after the fact.
Read the case →
API reliability
2026
Harwatch
Freeze an API from a HAR. Fail the build when a field disappears.
Outcome: API drift becomes a red build, not a customer report.
Read the case →
Documentation · Trust
2026
mimicplus Docs
Public notes for mimicplus. No keys. No customer recordings.
Outcome: Documentation a reader can trust because it shows its limits.
Read the case →