Tejas Penmetsa

About

Built and deployed a production LLM/RAG document automation system for business documents (contracts/claim forms) that extracts schema-validated JSON, generates grounded summaries/Q&A, and integrates into transaction systems via APIs. Emphasizes real-world reliability: hallucination controls, layout-aware parsing with OCR fallback, Step Functions-orchestrated workflows with retries/timeouts, and human-in-the-loop review designed in close partnership with operations and claims stakeholders.

Experience

Education

Key Strengths

• Built and deployed production LLM document automation pipeline end-to-end
• Strong hallucination mitigation via schema validation, confidence thresholds, and verification pass grounded in source text
• Robust handling of messy real-world documents (scanned PDFs, tables, inconsistent layouts) using layout-aware chunking and table normalization with OCR fallback
• Production reliability engineering (monitoring, logging, retries, timeouts, failure isolation, human-in-the-loop)
• Orchestrated multi-step ML/LLM workflows with Airflow/Prefect/AWS Step Functions for scalable, debuggable operations
• Measurement-driven evaluation approach (ground-truth datasets, accuracy/completeness/hallucination metrics, latency/error tracking, A/B tests)
• Effective cross-functional collaboration translating ops/claims needs into concrete extraction requirements and review thresholds