Experience

Contribute to a Flask/MongoDB Unified Verb Index, serving VerbNet, PropBank, FrameNet, and OntoNotes to NLP researchers; run 2 production Linux servers at 99.99% uptime over 18,772 health checks with zero-downtime gunicorn reloads under systemd.
Rebuilt ingest as a Python ETL pipeline: NLTK parses FrameNet, spaCy dependency parses generate syntactic frame diagrams, and cross-links across 4 corpora (3K+ entries) are computed once into denormalized collections, replacing request-time joins with single-document reads.
Cut p95 search and page latency from 897ms to 287ms across 20+ views by replacing full collection scans with MongoDB indexes and moving repeated Jinja2 lookups into precomputed fields; patched OpenSSL CVEs and upgraded to Flask 3 / Python 3.12, gated by pytest in GitHub Actions CI.