Meghavardhan Ketireddi

About

Built a production GPT-4/LangChain/Pinecone RAG “AI Copilot” at Northern Trust to automate financial report generation and analyst Q&A over internal structured (SQL warehouse) and unstructured policy data. Focused on real-world production challenges—grounding and latency—achieving major speed gains (seconds to milliseconds) via MiniLM embedding optimization and Redis caching, and implemented rigorous testing/evaluation with MLflow-backed metrics while aligning compliance and finance stakeholders for deployment.

Key Strengths

• Built and deployed a GPT-4 RAG copilot for financial analysts using SQL + unstructured policy docs
• Improved retrieval accuracy via semantic chunking and metadata tagging
• Reduced latency from seconds to milliseconds through embedding optimization and caching
• Switched from large OpenAI embeddings to domain-specific MiniLM, cutting vector size ~40% without losing semantic accuracy
• Designed reliable agent workflows with modular/unit testing plus synthetic eval sets for accuracy/grounding/latency
• Production monitoring of LLM system quality using MLflow and metrics (precision/recall, response rate)
• Orchestrated ETL and model retraining with Airflow; multi-agent LLM workflows with LangGraph
• Effective cross-functional communication with compliance/finance using diagrams and clear analogies to secure production approval

Experience

Education

Skills

Languages

Certifications