Applied AI / RAG Product
Sole-built job-discovery and application pipeline that ingests live postings, ranks them against an embedded resume/profile corpus, and drafts tailored application material. The system moves from regex pre-filtering to local Ollama embeddings, Claude Haiku triage, Claude Sonnet drafting, and pgvector-backed retrieval over a structured personal profile. It has async ingestion, migrations, structured logging, cost tracking, and a FastAPI + HTMX tracker UI around the model layer.
Python, Anthropic API, Claude Haiku, Claude Sonnet, Ollama, nomic-embed-text, Postgres 16, pgvector, FastAPI, HTMX, Alembic, Typer, psycopg3 async, RAG
Decision Systems
Live-money multi-strategy runner against Interactive Brokers ForecastEx using ib_async, with a hard daily loss circuit-breaker and an 8-point trade-approval gate before any prospective entry. The interesting part is not finance as an aesthetic; it is automation under real risk, where ranking, approval logic, scheduling, ledgering, and shutdown behavior all have consequences.
Python, ib_async, IBKR TWS/Gateway, ForecastEx, Postgres trade ledger, APScheduler, risk circuit breaker, approval gates
Explainable Forecasting / Defence AI
Led engineering for Vimy's DND IDEaS Fast Forward project forecasting global emerging threats using a hybrid neural-network and hidden-Markov-model approach. Built and presented an explainable conflict-forecasting concept live to a technical and military review audience, tying public information, statistical modeling, and machine-learning forecasts into a decision-support workflow.
Hybrid neural networks, Hidden Markov models, Explainable AI, Conflict forecasting, DND IDEaS, Decision support, Policy-aware outputs