Computer Science · Applied AI · Data & Reliability Engineering
Building evidence-driven AI, data, developer tooling and distributed systems with a focus on evaluation, reproducibility, operational safety and production engineering.
Each selected repository is structured so a reviewer can understand the problem and engineering thesis in roughly 60–90 seconds, run a local proof without private credentials where practical, and then follow a short path into architecture, tests and evidence.
Public claims deliberately distinguish implemented, tested, demonstrated and live/production-verified behaviour rather than treating them as equivalent.
| Project | Focus |
|---|---|
| ActionSeam | Adversarial conformance testing for agent runtimes, action boundaries and committed effects |
| Oxigênio Brasil | PT-BR ML/LLM evaluation, reproducible benchmarks, data governance and evidence |
| RepoOps | Bounded software maintenance, verification evidence and independent acceptance |
| Debug Evidence | Read-only incident diagnosis, local evidence collection, redaction and reproducible bundles |
| Credential Auditor | Reachable Git-history credential hygiene with redacted fail-closed reports |
| ChainEvidence | Reorg-aware EVM indexing, transactional recovery and verifiable on-chain data lineage |
| WM3 Request Protocol | Versioned request contracts with source of truth, scope, acceptance and evidence |
Handle This — outcome-oriented personal operations agent built around persistent Cases, guarded representation approvals, scheduled follow-up and fresh outcome verification. The Product Gold repository remains private; a sanitized reviewer build can be shared on request.
- executable evidence before broad claims;
- negative controls and fail-closed states;
- explicit authority and scope boundaries;
- reproducible identities, digests and provenance;
- separation of planning, execution, verification and acceptance;
- public claims bounded to the exact mechanism actually tested.
Most repositories are intentionally experimental and document what they do not prove as carefully as what they do.