🎥 Demo & Walkthrough: LinkedIn post

Note: The raw data and initial EDA notebook are not included in this repository due to confidentiality restrictions.

This repository contains the codebase for GrandView, a fully functional, lightweight AI app built during an in-company project with Prosol (Grand Frais). The goal was to create a Strategic Product Copilot that transforms raw transactional data into tailored business strategies using:

• Product Segmentation (via unsupervised clustering)
• Retrieval-Augmented Generation (RAG)
• Large Language Models (LLMs)

This tool is designed for retail teams to gain strategic clarity and act on insights — all within a Streamlit app that runs locally.

(Note: Raw data and full EDA files are not shared in this public repository)

• Performed advanced feature engineering from offline retail transactional data
• Applied unsupervised clustering to build segmentation categories (e.g., timing, loyalty, payment behavior)
• Annotated cluster outputs with interpretable product labels to aid LLM reasoning

• Generates product-level strategies using segmentation labels, sales metrics, and prompting techniques
• Powered by FAISS vector search and Gemini/Mistral LLMs

• Handles open-ended questions like "Which products need better retention?"
• Implements RAG to retrieve similar product profiles and injects them into LLM prompts

• Collects business expert feedback
• Embeds a feedback-memory loop to continuously improve LLM responses
• Fine-tuning ready

• Enables download of filtered product lists by segmentation label and product hierarchy (category, family, sub-family)
• Useful for campaign planning or strategic analysis

• Clear cache, reset files, and manage app stability in one place

• Gemini API and Mistral 7B via Hugging Face Transformers

• Fully functional without cloud infra – built entirely on a local laptop
• Designed to be adopted by business users without technical training
• Modular design with future integration potential (e.g., Store Review System, Forecasting Modules)

• Integrate with Prosol’s Store Review (POS Insight) platform
• Connect teammate modules: demand forecasting, price elasticity, and customer segmentation
• Expand into a fully agentic decision-support system

Special thanks to the Prosol Data team and professors at emlyon business school for guidance and feedback throughout the project.