Hiya Kothari

About

Built a Python-based geospatial machine learning backend for PFAS contamination risk mapping, including reproducible feature pipelines, ensemble modeling, and a FastAPI layer for visualization/analysis. Emphasizes data integrity and robustness (CRS/coverage checks, fail-fast validation) and has led safe backend refactors using feature flags, idempotent backfills, and Postgres RLS for secure, queryable results delivery.

Experience

Education

Key Strengths

• Designed and evolved geospatial ML backend pipeline for PFAS contamination risk mapping
• Strong focus on reproducibility via versioning, deterministic preprocessing, and experiment parameter/schema logging
• Data quality and integrity engineering (validation checks, quality reports, fail-fast pipelines)
• API performance optimization (cursor-based pagination, indexing, payload limits, moving expensive logic out of request path)
• Safe migration/refactor execution (feature flags, parallel runs, incremental rollout, easy rollback)
• Production-grade security implementation (OAuth2/JWT, role-based access, Postgres RLS policies and testing)
• Improved model accuracy to ~78% and enabled geospatial insights supporting policy discussions
• Shipped AI resume-to-job matching feature end-to-end (embeddings + vector DB + AWS deployment)
• Data-driven iteration using implicit user feedback signals (clicks/skips/manual selections)
• Balanced model tradeoffs via A/B testing across relevance, latency, and cost
• Improved matching precision by 10% and reduced response time by 30% through re-ranking and skill weighting
• Production RAG/LLM pipeline design with robust guardrails (confidence thresholds, PII filtering, fallbacks)
• Empirical threshold calibration using labeled validation sets (precision@k >85% cutoff)
• Built fault-tolerant ingestion pipeline for messy resumes with retries, DLQs, and observability
• Designed continuous LLM eval loops (grounding/hallucination metrics, human review, regression prevention) and drove measurable grounding improvements (+15%)