Applied AI & Software Engineer. I build analytical runtimes, AI agents, agent infrastructure and local-first AI tooling, and I measure whether they work.
| Project | What it is | Strongest evidence |
|---|---|---|
| BI Notebook Lab | Browser-based analytical runtime that teaches how BI semantic models compute: DAX lexer → parser → binder → evaluator, filter context, relationships, Power Query | 1,024 tests · 83-case DAX conformance suite with documented divergences · 100k-row benchmark |
| OpsTwin | Operational simulation lab for testing service-workflow changes with paired simulation, sensitivity and uncertainty ranges | 247 backend + 172 frontend tests · strict typing · live deployment |
| DataBrief AI | Bounded analytics workflow: CSV/XLSX → profiling → routed plan → sandboxed execution → grounded report | 176 backend tests · every finding cites the artifact it came from |
| Open VS Code Agent | Coding agent designed and benchmarked for a local 7B open-weight model, with explicit tools, verification and crash-safe mutation | 24-task benchmark: 37.5% → 75.0% grounded success after evidence-driven orchestration |
| JARVIS OS | Personal AI operating system built around one question, what needs my attention today?, with specialist agents, governed actions and human approval | Evidence-required attention · autonomy ladder · verified execution and recovery |
| HALO Control | Control plane for a local AI workstation: model registry, capability routing, telemetry, workload authorization | Deterministic routing · metadata-only activity log · explicit mocked/real evidence classes |
JARVIS OS, Open VS Code Agent and HALO Control are public architecture editions of private systems. BI Notebook Lab, OpsTwin and DataBrief AI are open source.
Progression: analytical systems (BI Notebook Lab, OpsTwin) → applied AI (DataBrief AI) → AI agents (Open VS Code Agent) → agent systems (JARVIS OS) → local AI infrastructure (HALO Control).
OpenLIT: LangGraph memory connector (open pull request). A LangGraph Store memory connector with memory CRUD/search, namespace mapping, authentication, safe content handling, docs and integration tests.
| AI systems | Agent orchestration (CrewAI), MCP, tool calling, memory, human-in-the-loop approval, Ollama and open-weight models |
| Evaluation | Benchmark harnesses, conformance suites, failure taxonomies, Vitest, Pytest, Playwright |
| Backend | TypeScript, Node.js, Fastify, Python, FastAPI, SimPy |
| Frontend | React, Next.js, Vite, React Flow, Recharts |
| Data | PostgreSQL, SQLite, IndexedDB, DAX and semantic modelling, Power BI concepts |
| Infrastructure | Local inference, Vercel, GitHub Actions |
Agent systems on local and open-weight models, with an emphasis on reliability, memory, evaluation and orchestration, and on analytical runtimes that explain their own results.