Yash Sanjay Zaveri

About

Built and deployed an AI Copilot for Healthful Telehealth that helps dietitians generate personalized meal plans using patient data and real-time clinical context. Stands out for owning the full lifecycle—from workflow discovery and ETL/RAG architecture to production incident response and post-launch stabilization—while delivering roughly 30% gains in retrieval accuracy and latency.

Key Strengths

• owned end-to-end AI copilot deployment from discovery through post-launch stabilization
• built production RAG workflows for healthcare use cases
• improved response latency and retrieval accuracy by around 30%
• strong stakeholder translation between dietitians and product/technical teams
• debugged non-deterministic LLM output issues across ingestion, retrieval, and prompting layers
• designed robust preprocessing for messy multi-source healthcare data
• implemented validation, fallback, retry, logging, and monitoring patterns for reliability
• optimized indexing, chunking, and query execution for real-time performance
• Shipped production LLM agent (dietitian AI copilot) with measurable impact (~30% performance improvement)
• Designs robust multi-step agent pipelines (retrieval, reasoning, validation, structured outputs)
• Strong reliability engineering for LLM systems (guardrails, retries/fallbacks, human-in-the-loop)
• Schema-first approach (strict JSON schema enforcement and validation for safe downstream use)
• Evaluation and monitoring discipline (offline edge-case tests + online metrics like latency, retry/fallback rates)
• Handles messy semi-structured operational data via normalization, defaults, and safe fallbacks

Reference Highlights

• Production-focused mindset
• Builds robust, predictable systems for real-world use
• Strong end-to-end ownership of agent workflows
• Thinks through edge cases and real-world complexity
• Strong handling of messy and incomplete data
• Methodical, structured, data-driven quality improvement approach
• Effective monitoring and closing the loop from metrics to iteration
• Good judgment in designing safe fallbacks
• Improved system performance by ~30%
• Reduced manual effort for dietitians and enabled scalable workflows

Experience

Education

Skills

Languages