Shreya Andela

About

Built and shipped a production LLM-powered RAG assistant for enterprise internal document search (PDFs, knowledge bases, structured data), addressing real-world issues like noisy documents, hallucinations, and latency with grounded prompting, retrieval-confidence fallbacks, and performance optimizations. Also partnered with compliance and business teams at JPMc to deliver a solution aligned with regulatory constraints, supported by monitoring, feedback loops, and systematic evaluation.

Experience

Education

Key Strengths

• Built and deployed production RAG system over enterprise PDFs/knowledge bases/structured data
• Improved retrieval quality for noisy/unstructured documents via structure-aware chunking and selective re-embedding
• Reduced hallucinations using strict context-grounded prompts and low-confidence fallback logic
• Production performance optimization (caching, top-K retrieval optimization, pipeline parallelization)
• Established monitoring and feedback loops to tune post-deployment performance
• Hands-on orchestration for ingestion/embedding/model update workflows using Airflow, Prefect, and Kubernetes-native workflows
• Reliability-focused evaluation approach with defined metrics (accuracy, latency, hallucination rate), unit/integration tests, offline labeled evals, and drift tracking
• Effective cross-functional delivery with compliance and business stakeholders for regulated internal AI assistant