At Figure 99, I reduced an inventory dashboard's p95 latency by 29% by profiling database queries, removing N+1 access patterns, and adding targeted indexes.
I also built extraction and validation for automated supplier-invoice ingestion using Python, Tesseract OCR, and structured LLM outputs. I evaluated 200 invoices and added financial reconciliation and confidence-based validation to keep uncertain records out of downstream inventory writes.
For a six-person UBC capstone, I established the FastAPI backend for Digital Work Artifact Miner, delivering most REST endpoints and a Python AST-based code-analysis engine. I built consent-gated LLM enrichment with privacy controls, structured parsing, fallbacks, and failure isolation.
I've also built geospatial routing tools with Next.js, PostgreSQL, Mapbox, and OpenStreetMap, and owned authentication and messaging interfaces for a real-time chat platform.
