Adarsh Vodnala

About

Built and shipped production AI features in the automotive dealership domain, including an end-to-end computer-vision damage detection system for trade-ins and a tool-calling, RAG-enabled LotSync AI Agent that answers inventory/VIN questions using strict schemas and internal APIs to avoid hallucinations. Also developed a Dagster + Oracle automated reporting pipeline as a Graduate Research Assistant, supporting 15+ university departments with normalized, reliable ETL workflows.

Experience

Education

Key Strengths

• Owned and shipped AI-driven vehicle damage detection end-to-end (photo upload → vision models → repair cost estimation → valuation)
• Balanced accuracy vs response time via rapid initial response with deeper follow-up analysis
• Improved AI transparency post-launch by adding repair types/cost breakdown and selectable repair options
• Built production LLM agent using tool-calling with strict schemas to prevent hallucinations and ensure responses come from internal APIs
• Implemented guardrails: parameter validation, conversation state management, logging, and tool-execution fallbacks
• Designed LLM eval loop with dealership query test set; improved tool selection by refining tool descriptions and schemas
• Built scalable ETL/reporting pipeline on Oracle using Dagster with normalization, data quality checks, retries/backoff, and Dockerized deployment
• Scaled reporting via batch scheduling, indexing, and caching; supported 15+ university departments