You can deploy this model and start earning money today!

eyuleo/
CreditRisk

Views

No views yet

CrediTrust Financial AI Complaint Chatbot

Project Overview

This project aims to develop an internal AI tool for CrediTrust Financial to analyze unstructured customer complaint data. Leveraging Retrieval-Augmented Generation (RAG), the tool will enable product managers and non-technical teams to quickly identify major complaint trends and get evidence-backed answers.

Business Objectives / KPIs

Decrease the time to identify major complaint trends from days to minutes.
Empower non-technical teams to get answers without needing a data analyst.
Shift the company from reacting to problems to proactively identifying and fixing them.

Project Structure

credittrust_complaint_chatbot/ ├── data/ ├── notebooks/ ├── src/ ├── vector_store/ ├── .gitignore ├── README.md └── requirements.txt

Setup and Installation

Clone the repository:
bash
1git clone <your-repo-url> 2cd credittrust_complaint_chatbot
Create a virtual environment (recommended):
bash
1python -m venv venv 2source venv/bin/activate
Install dependencies:
pip install -r requirements.txt

How to Run

Task 1: Exploratory Data Analysis and Data Preprocessing

Using Jupyter Notebook (for interactive exploration):
jupyter notebook notebooks/eda_preprocessing.ipynb
Running as a script:
python src/eda_preprocessing.py
(Ensure your raw complaints.csv is in the data/ directory or update the script path.)

Task 2: Text Chunking, Embedding, and Vector Store Indexing

Run the script:
python src/chunk_embed_indexing.py
(This script will read data/filtered_complaints.csv and save the vector store to vector_store/.)

Task 3: Building the RAG Core Logic and Evaluation

Purpose: To implement the core Retrieval-Augmented Generation (RAG) pipeline, which retrieves relevant complaint data and uses an LLM to generate answers. This script also performs a qualitative evaluation.

Run the script:

python src/rag_application.py

(This script will load the vector store, initialize the LLM, and run a series of evaluation queries, printing results to the console.)

Note on API Quota:
If you encounter ResourceExhausted errors, your Google Gemini API free-tier quota might be temporarily exceeded. Please wait and try again later, or consider upgrading your Google Cloud project.

Task 4: Creating an Interactive Chat Interface

Purpose: To provide a user-friendly web interface for interacting with the RAG system, allowing non-technical users to ask questions and view evidence-backed answers with source traceability.

Run the Gradio application:

python app.py

eyuleo/CreditRisk