RAG Complaint Chatbot

An end-to-end Retrieval-Augmented Generation (RAG) chatbot for consumer complaints.
The system processes raw CFPB complaint data, builds embeddings, and provides natural language responses via a RAG pipeline.

---

Table of Contents

• Project Overview
• Business Context
• Objectives
• Dataset Overview
• Project Structure
• Architecture
• Pipeline & Modeling Approach
• MLOps & Engineering Practices
• Setup & Installation
• Running the Project
• Technologies Used
• Author

---

Project Overview

This project implements a RAG chatbot for consumer complaints.
It enables efficient information retrieval from large complaint datasets while generating coherent answers in natural language.

The system covers:

• Data ingestion and cleaning of CFPB complaint datasets
• Text chunking and embedding generation
• FAISS/Chroma vector store construction
• RAG pipeline for retrieval + generation
• Qualitative evaluation of responses
• Streamlit/Gradio user interface
• Containerized deployment with Docker
• Data versioning and reproducibility with DVC

---

Business Context

Financial institutions and consumer protection agencies need fast, accurate responses to complaints.
Manual review is slow, error-prone, and expensive.

This chatbot supports:

• Automated complaint responses
• Insight extraction from large datasets
• Research and monitoring of consumer issues
• Enhanced customer support efficiency

---

Objectives

• Build a retrieval-augmented system for complaint data
• Generate accurate and context-aware responses
• Maintain a scalable vector store for embeddings
• Ensure reproducibility with DVC and versioned datasets
• Deploy via API or interactive UI

---

Dataset Overview

Source: CFPB consumer complaints

Key fields:

Column	Description
ComplaintID	Unique complaint identifier
Product	Complaint category
Issue	Detailed complaint issue
ConsumerComplaint	Complaint narrative
Company	Company involved
DateReceived	Complaint submission timestamp

Derived features:

• Cleaned and normalized complaint text
• Chunked complaint segments for embedding
• Metadata for retrieval and filtering

---

Project Structure

rag-complaint-chatbot/
│
├── config/                      # YAML configs: paths, embeddings, RAG, UI
├── data/                        # Raw, interim, processed, external datasets
├── vector_store/                # Persisted FAISS / Chroma embeddings
├── notebooks/                   # EDA, preprocessing, embeddings, evaluation
├── scripts/                     # CLI scripts to run pipelines and UI
├── src/
│   └── rag_chatbot/             # Main Python package
│       ├── core/                # Settings and configuration loader
│       ├── data/                # Data loading and filtering
│       ├── preprocessing/       # Text cleaning & normalization
│       ├── chunking/            # Text chunking logic
│       ├── embeddings/          # Embedding generation
│       ├── vectorstore/         # FAISS / ChromaDB wrappers
│       ├── rag/                 # RAG pipeline (retriever + generator)
│       ├── evaluation/          # Qualitative evaluation logic
│       ├── ui/                  # Streamlit / Gradio interface
│       └── utils/               # Shared utilities
├── tests/                       # Unit & integration tests
├── docker/                       # Dockerfile and compose
├── dvc.yaml                       # DVC pipeline
├── params.yaml                    # Global parameters
├── requirements.txt
├── pyproject.toml
└── README.md

---

Architecture

---

Pipeline & Modeling Approach

1. EDA – Explore distribution of complaint categories, narrative length, missing values.
2. Preprocessing – Text cleaning: lowercasing, punctuation removal, stopword handling.
3. Chunking – Split long complaint narratives into manageable text chunks.
4. Embedding – Generate vector representations for each chunk.
5. Vector Store – Store embeddings in FAISS or Chroma for efficient retrieval.
6. RAG Pipeline – Combine retrieval and generation to answer queries.
7. Evaluation – Qualitative analysis of responses against test complaints.
8. UI / API – Interactive interface for testing the chatbot or integrating into systems.

---

Engineering Practices

• Data Versioning: DVC tracks raw and processed datasets, embeddings, and vector stores
• Containerization: Docker & docker-compose for easy deployment
• Testing: Unit and integration tests for preprocessing, chunking, embedding, and RAG pipeline

---

Setup & Installation

Clone the repository:

git clone https://github.com/<username>/rag-complaint-chatbot.git
cd rag-complaint-chatbot

Create a virtual environment:

python -m venv .venv
source .venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Or install in editable mode:

pip install -e .

---

Running the Project

Run full pipeline (DVC)

dvc repro

Launch UI

python app.py

Technologies Used

• Python 3.10+
• OpenAI API / LangChain
• FAISS / Chroma
• DVC (data version control)
• FastAPI
• Streamlit / Gradio
• Docker & docker-compose
• Pytest for testing

---

Author

Tibebu Kaleb – Full-stack AI/ML engineer with experience in NLP and RAG pipelines