A Retrieval-Augmented Generation (RAG) system designed to help analysts and customers query consumer financial complaints from the CFPB dataset using natural language. Built with LangChain, ChromaDB, Sentence Transformers, and Streamlit.

• End-to-end pipeline from raw complaint data to interactive semantic search.
• Data cleaning, narrative filtering, and chunking for optimized embeddings.
• Vector search with ChromaDB using metadata for traceability.
• Question-answering using MiniLM for embedding and Mistral-7B-Instruct for LLM generation.
• Clean, interactive chat interface built with Streamlit.
• Modular design with reproducible scripts and Docker support.

rag_project/
├── app.py                         # Streamlit interface (Task 4)
├── Dockerfile                     # Docker container setup
├── requirements.txt               # Python dependencies
├── data/
│   └── filtered_complaints.csv    # Cleaned data ready for chunking
├── vector_store/
│   └── chroma_db/                 # Persistent ChromaDB store
├── notebooks/
│   └── eda_preprocessing.ipynb    # Task 1: EDA + cleaning notebook
├── src/
│   ├── chunking_embedding_indexing.py  # Task 2: chunk + embed + index
│   ├── rag_pipeline.py            # Task 3: retrieval + generation logic
│   └── utils.py                   # Optional helpers
└── .dockerignore

pip install -r requirements.txt

jupyter notebook notebooks/eda_preprocessing.ipynb

python src/chunking_embedding_indexing.py

python src/rag_pipeline.py

streamlit run app.py

docker build -t creditrust-rag .

docker run -p 8501:8501 creditrust-rag

The app will be available at http://localhost:8501

Check evaluation.md or the final report for:

• A table of 5–10 test questions
• Quality scores from 1–5
• Example retrieved chunks
• Generation analysis

• CFPB Public Complaint Dataset
• Hugging Face Transformers
• LangChain
• ChromaDB
• Streamlit

• Add support for feedback-driven retraining
• Swap in larger LLMs via LangChain integration (e.g., Llama 3)
• Add support for citation highlighting
• Deploy as a cloud API