ATB – Local-first audit trail for AI systems

Built ATB as a local-first audit trail for privacy-sensitive AI workflows, recording AI events as tamper-evident bundles that can be inspected locally and exported as deterministic evidence. Implemented SHA-256 hash chains over RFC 8785 canonical JSON to detect mutation, reordering, and deletion in recorded traces. Added AES-256-GCM client-side encryption and deterministic export paths for incident review, customer handoff, and internal audit and privacy review workflows. Developed a local viewer and dashboard with timeline, graph, inspector, and privacy reveal audit logging for sensitive data access. Shipped tracing middleware for LangChain in Python and Vercel AI SDK in TypeScript so teams can instrument agents and copilots without relying on hosted observability platforms.​