Medhovarsh Bayyapureddi

About

AI/ML-focused backend engineer who shipped two production systems: PersonaPal (agentic LLM chatbot with RAG, FAISS-based retrieval, and Redis semantic caching) and CervixScan (clinical diagnostics platform with PostgreSQL data modeling and human-in-the-loop safety for low-confidence predictions). Demonstrates strong performance/reliability work (indexed vector search, caching, query optimization to ~200ms) and end-to-end ownership from orchestration design through deployment.

Experience

Education

Key Strengths

• Built production backend for agentic AI chatbot (Python/FastAPI) with RAG and session endpoints
• Performance optimization of vector retrieval (migrated to FAISS indexed search) to reduce latency at scale
• Implemented Redis semantic cache using embedding similarity threshold to cut repeated-query latency and API cost
• Designed persona orchestration and reinforcement layer to prevent context drift and maintain consistent tone
• Designed plan-then-execute agent workflow with step verification and retry/escalation/stop behavior
• Production engineering practices: Pydantic typing, async I/O, Celery+Redis for long-running inference, Poetry/Conda env reproducibility, Pytest, structured logging with p99 latency tracking
• PostgreSQL clinical data model (Patients/Scans/Findings) with constraints; improved slow report queries via composite indexes and EXPLAIN ANALYZE (reduced to ~200ms)
• Operational guardrails for AI in healthcare: monitoring confidence scores, evaluation pipelines, and human-in-the-loop fallback for low-confidence predictions