Shreyas Krishnareddy

About

Backend/AI engineer who built and refactored a production RAG system over IRS Form 990 filings for 60 nonprofits, using a dual-path architecture (deterministic financial ranking + TF-IDF semantic retrieval) to keep latency sub-2s and reduce hallucinations. Demonstrates strong API craftsmanship in FastAPI (contract-first, OpenAPI-driven) plus production-grade security for multi-tenant systems (JWT, RBAC, Supabase-style RLS) and careful migration practices (feature flags, traffic mirroring, incremental rollout).

Experience

Education

Key Strengths

• Designed and built a dual-path RAG backend (deterministic ranking + TF-IDF) optimized for correctness and speed
• Strong focus on interpretability and predictable behavior (avoided vector DB for small dataset; added thresholds and fallbacks)
• Reduced hallucinations and improved trust via grounded responses and explicit failure handling
• Robust data integrity approach: normalized inconsistent IRS XML into validated JSON schema with safe defaults
• API design for frontend enablement: contract-first FastAPI with Pydantic/OpenAPI, versioning, consistent errors, and conventions
• Production reliability practices: async where needed, timeouts/retries, caching, health checks, structured logging, and metrics
• Risk-managed migrations: feature flags, mirrored traffic, incremental rollout with monitoring before deprecating old path
• Security depth: server-side JWT verification, scoped access (OAuth/API keys), RBAC, and database-enforced multi-tenant RLS