Walmart Global Tech – RAG Chatbot

Overview

Built a retrieval-augmented chatbot to help internal teams find accurate answers across enterprise knowledge sources. The work emphasized retrieval quality, grounding, and reliability under real constraints like document freshness, ambiguity, and practical evaluation needs. Because parts of the work are confidential, this write-up focuses on architecture, tradeoffs, and what I learned from iterating on retrieval quality rather than proprietary details.

Highlights

• Designed a pragmatic RAG pipeline with retrieval quality as the success metric, not just model output.
• Iterated on retrieval and prompting to improve relevance and reduce unsupported responses.
• Built an evaluation mindset into the workflow with repeatable tests to measure improvements over time.
• Documented architecture and tradeoffs in a way that can be shared safely without exposing confidential data.

Tech

Python, RAG, embeddings, vector search, evaluation-driven iteration