Yash Amre

About

Built and deployed a production LLM-powered internal AI assistant using a RAG pipeline to help teams search internal PDFs/knowledge bases and generate grounded summaries/answers. Demonstrates strong end-to-end ownership (ingestion through APIs) plus production rigor (monitoring/logging/CI-CD, evaluation metrics) and practical optimizations for hallucination, latency, and answer quality (thresholding, fallbacks, caching, async, re-ranking, two-tier model routing).

Experience

Education

Key Strengths

• Built and deployed a production RAG-based internal AI assistant for querying internal documents
• End-to-end ownership: ingestion, embeddings, vector search, prompt design, and API development
• Hallucination reduction via strict context grounding, similarity thresholds, constrained prompts, and low-confidence fallbacks
• Latency reduction via chunking optimization, context limiting, caching, and async requests
• Production reliability practices: step-level error handling, retries/timeouts, structured logging, metrics/alerts, and CI/CD
• Evaluation-driven iteration using offline test sets, unit tests, and production monitoring (accuracy, retrieval relevance, latency, failure rate, cost)
• Improved system performance with two-tier model strategy and retrieval re-ranking
• Effective collaboration with non-technical stakeholders by aligning on success metrics and iterating via demos/feedback