Portfolio
Systems built, documented, and reflected upon.
AI Engineering
A 3-agent CrewAI pipeline running on local LLM inference (Qwen 2.5 Coder 14B via Ollama) that reads real filesystem data, writes and executes Python code with binary validation, and produces portfolio-ready HTML dashboards — orchestrated through a Streamlit Mission Control GUI at zero API cost.
Data Engineering
Automated schema migration system that validates every change against a registry of known queries before execution — catching breaking changes that schema-level analysis misses, with zero-downtime deployment via shadow-table swaps and concurrent index operations across a 47-table PostgreSQL database.
AI Engineering
RAG pipeline that transforms 2,000+ workforce education documents into a semantic knowledge base — with structure-aware chunking, hybrid retrieval (BM25 + cosine similarity), and LLM-verified answer synthesis achieving 89% retrieval relevance on domain-specific queries.
Infrastructure
Unified observability stack combining Prometheus metrics, Loki log aggregation, and Grafana dashboards — replacing SSH-and-tail debugging with correlated telemetry, domain-specific alerting, and 91% faster root cause identification across 4 microservices.
Systems Architecture
A modular workflow engine that models operational processes as dependency graphs — with typed task contracts, container-isolated execution, automatic rollback, and structured audit logging that eliminates stale documentation by making the code the process definition.
Data Engineering
Event-driven ingestion pipeline that unifies 4 source platforms through Kafka, processes streams with Go microservices, and delivers sub-second analytical queries via ClickHouse — replacing 12-hour batch delays with real-time operational visibility.
AI Engineering
A 3-agent CrewAI pipeline coordinated by a state machine that reads real data, generates validated Python code, and produces portfolio-ready HTML dashboards — with structured retry logic and output schemas enforced at every handoff.