Roshan Erukulla

About

Built and deployed a production LLM-powered RAG assistant for healthcare teams (care managers/support) to answer questions from clinical and policy documentation, emphasizing trustworthiness via improved retrieval, reranking, and strict grounding prompts to reduce hallucinations. Also has hands-on orchestration experience with Apache Airflow for end-to-end ETL/ML workflows and applies rigorous testing/metrics (hallucination rate, tool-call accuracy, latency, cost) to ensure reliable AI agent behavior.

Experience

Education

Key Strengths

• Built and deployed a production LLM RAG system for clinical/policy document Q&A
• Strong focus on factual grounding and reducing hallucinations in healthcare use cases
• End-to-end pipeline ownership (ingestion, chunking, embeddings/indexing, retrieval/reranking, generation, deployment)
• Production deployment experience (FastAPI service, Dockerized)
• Workflow orchestration for ML/data pipelines using Apache Airflow
• Reliability engineering for AI agents (guardrails, schema validation, confidence thresholds, fallbacks)
• Testing and evaluation discipline (scenario/regression tests; tracks success, hallucination rate, tool accuracy, latency, cost)
• Pragmatic model/retrieval/prompt selection based on accuracy, latency, cost, compliance, explainability tradeoffs
• Effective communication with non-technical stakeholders using business metrics and risk framing