Pooja Murigappa

About

ML/LLM engineer at Charles Schwab who built a production loan-advisor chatbot integrated with internal knowledge and loan-calculator APIs, adding strict numeric validation to prevent rate hallucinations and optimizing context to control costs. Also runs ~40 Airflow DAGs orchestrating retraining/ETL/drift monitoring with an automated Snowflake→SageMaker→auto-deploy pipeline, and uses rigorous testing plus canary rollouts tied to business metrics and compliance constraints.

Experience

Education

Key Strengths

• Built and deployed a production loan-advisor LLM chatbot at Charles Schwab to reduce repetitive loan officer inquiries
• Implemented hallucination guardrails by validating numeric/rate outputs against authoritative rate tables before responding
• Reduced LLM API costs by limiting context with a sliding window (last 5 exchanges) after excessive spend (~$500/day)
• Operates Airflow at scale (≈40 production DAGs) for retraining, ETL, and drift monitoring, replacing unreliable cron jobs
• End-to-end automated ML pipeline: Snowflake data pull → SageMaker training → metric validation → auto-deploy on improvement
• Strong evaluation discipline: unit + integration tests using historical tickets, plus staged rollout (10% traffic) and escalation-rate comparisons
• Pragmatic model strategy balancing latency and accuracy (sub-3s constraint; GPT-3.5 for chat, fine-tuned BERT for classification)
• Effective non-technical stakeholder management using business-metric demos and compliance boundary-setting (green/red capability chart)