How ridabayi/flipkart-product-recommender works
Project type and purpose
• This is a full-stack web application that:
– Ingests and preprocesses Flipkart product review data
– Builds/runs a Retrieval-Augmented Generation (RAG) chain using an LLM
– Serves product recommendations via a web UI
• Key indicators: templates/ + static/ directories (Flask front end), Flipkart/ package (backend logic), Data/flipkart_product_review.csv (raw data).
File structure
Top-level directories/files:
• Data/
– Raw CSV of reviews
• Flipkart/ (backend core)
– Data_ingestion.py (loads raw data)
– Data_converter.py (processes/embeds data)
– rag_chain.py (LLM + vectorstore orchestration)
– config.py (app configuration)
– vault_api_keys.py (securely fetches secrets)
• static/, templates/
– Frontend assets (CSS, HTML)
• utils/
– custom_exception.py, logger.py (shared utilities)
• setup.py, requirements.txt (packaging + dependencies)
Key insights from README
• You must infer architecture from code and folders
a. Frontend
– Browser → HTML (templates/index.html) + CSS (static/style.css)
– Sends user queries (e.g., “recommend me…”) via HTTP
b. Backend (Flask application)
– API endpoints (e.g., /recommend)
– config.py + vault_api_keys.py for environment & secrets
– utils/logger.py & utils/custom_exception.py
c. Data Pipeline
– Data_ingestion: reads Data/flipkart_product_review.csv
– Data_converter: text cleaning, embedding generation
– Vector Store (e.g., in-memory or disk-based—explicit or abstract)
d. RAG Chain (rag_chain.py)
– Retrieval: query vector store for nearest neighbors
– LLM call: passes context + prompt to external LLM service (OpenAI or similar)
e. External Services
– Vault or secret manager (for API keys)
– LLM provider (OpenAI, Azure OpenAI, etc.)
Relationships and interactions
• User → Frontend (HTTP GET/POST)
• Frontend → Flask Server (HTTP request)
• Flask Server → Data Pipeline / RAG Chain (in-process calls)
• RAG Chain → Vector Store (local or remote DB)
• RAG Chain → External LLM API → returns generated recommendations
• Flask Server → Frontend (HTTP response with recommendations)
Architectural patterns/design principles
• Separation of concerns:
– Presentation layer (templates/static)
– Business logic (Flipkart/ modules)
– Data access & processing (Data_ingestion, Data_converter)
• Modular design: easy to swap out vector store or LLM provider
• Utility layer for cross-cutting concerns (logging, exception handling)
• Package structure supports pip install . (setup.py + .egg-info)
Detail
• Every Python module as a component or sub-component
• Both data storage (CSV, vector store) and external dependencies