Keerthana Priya

About

At KPMG, built and productionized a secure RAG-based LLM assistant that lets business and risk stakeholders query data warehouses in natural language, reducing dependence on data engineers for ad-hoc analysis. Demonstrates strong production rigor (Airflow orchestration, CI/CD, containerization), retrieval/embedding tuning (rechunking, semantic abstraction for structured data), and reliability controls (confidence thresholds, refusal behavior, monitoring and canary evals).

Experience

Education

Key Strengths

• Built and deployed a production LLM/RAG system enabling natural-language access to warehouse data
• Strong hallucination-mitigation approach using retrieval-only answering and confidence-based refusal
• Deep practical understanding of retrieval quality tradeoffs (chunking/embedding tuning) and structured-data embedding via semantic abstraction
• Production orchestration experience with Airflow for ingestion, validation, embedding regeneration, index refresh, and evaluation
• Designs reliable pipelines with idempotent tasks, checkpoints, and modular/reusable workflow components
• Metrics-driven evaluation and monitoring (hallucination rate, rejection rate, drift, latency, cost) with canary rollouts
• Effective collaboration with non-technical stakeholders to translate business needs into prototypes and iterate on prompts/retrieval/thresholds