Work on AI assurance, verification-guided reasoning, and the infrastructure underneath both. Featured items appear first.
Independent assurance and evidence platform for AI agents. Agent decisions arrive over OpenTelemetry or a one-line HTTP contract and are held in independent, append-only, hash-chained custody, making any later edit or deletion detectable. Krynix normalises them into one evidence model across frameworks and sits above whatever enforcement stack you already run. An optional transparent MCP gateway emits decisions with no agent-side code changes.
A direct implementation of the MSc dissertation research. An iterative controller couples a generator with typed verifiers across arithmetic, symbolic algebra via SymPy, and formal logic. Supports OpenAI and Anthropic APIs plus local inference via Ollama. Dockerized and Kubernetes-ready.
Fixed a URL resolution bug causing ECONNREFUSED and a compound resource ID bug causing 400 errors in the Azure DevOps community plugin. Added 4 regression tests and submitted a pull request with changeset.
ClickHouse-backed event pipeline with Python and Node SDKs, drift and fairness monitoring, RBAC, OIDC SSO, Prometheus metrics, and multi-channel alerting.
Machine learning pipeline for financial fraud detection with feature processing, evaluation, and benchmarking to reduce false positives.