Shanmukha Jayavarapu

About

Built and deployed a production LLM-powered calorie-counting chatbot that turns plain-English meal descriptions into normalized food entities, quantities, and calorie estimates using a hybrid transformer + rule-engine pipeline. Emphasizes reliability with schema/constraint guardrails, confidence-based routing (including embedding similarity search fallbacks), and strong observability/metrics (hallucination rate, calibration, latency, cost). Partnered closely with nutritionists to encode domain standards into mappings and validation logic.

Experience

Education

Key Strengths

• Built and deployed a production LLM-powered calorie counting chatbot converting free-text food logs into structured nutrition data
• Hybrid architecture combining transformer NLP with deterministic rule-based validation for safety/correctness
• Strong guardrails design (schema/type checks, unit normalization, realistic bounds) to mitigate probabilistic outputs and hallucinations
• Orchestrates multi-stage LLM workflows with modular components, observability, and error isolation
• Latency/cost optimization via confidence-based conditional routing and fallbacks (skipping heavy retrieval when high confidence)
• Metrics-driven evaluation across model and system KPIs (hallucination rate, calibration, latency, cost per request, drift monitoring)
• Effective cross-functional collaboration translating nutritionist domain expertise into validation rules and entity mappings