Abstract
The rapid growth of online retail has created enormous volumes of unstructured product data that most businesses struggle to convert into actionable intelligence. Customer reviews, product descriptions, pricing information, and rating histories collectively represent a rich source of market knowledge, yet extracting meaningful insights from this data at scale remains a persistent challenge for organizations operating in competitive e-commerce environments. This thesis presents ProductIQ, an intelligent analytics platform that addresses this gap by enabling users to submit natural language business queries and receive structured, evidencegrounded analytical reports derived from a corpus of 200,000 Amazon Electronics product records. The system retrieves semantically relevant products using BGE-M3 embeddings and ChromaDB vector search, then conditions Claude Opus 4.6 to synthesize the retrieved evidence into coherent analytical narratives covering key market findings, product comparisons, and strategic recommendations. By grounding every response in retrieved product data, the system reduces the risk of hallucination while producing outputs that reflect the actual state of the product catalog. Beyond the core retrieval and generation components, the platform incorporates a production-grade MLOps pipeline comprising MLflow experiment tracking, Docker containerization, and GitHub Actions for continuous integration and delivery. Evaluation across eight quantitative metrics confirms the quality of system outputs, with a BERTScore F1 of 0.9131 indicating strong semantic alignment with human-authored reference insights and a faithfulness score of 0.5567 reflecting meaningful grounding in retrieved evidence. All sixteen automated unit tests pass, confirming system reliability. This research demonstrates how retrieval-augmented generation can be applied to transform large-scale e-commerce product data into accessible, reliable business intelligence.