README.md

Fintech AI Agent

In this tutorial, we'll develop an intelligent system to process loan and insurance queries using AI agents.

1. Extract Loan Intent: Use LLM to classify user queries for loan intent (e.g., home improvement, medical) and predict eligibility with a RandomForest model.
2. Generate Insurance Claims: Simulate realistic auto insurance claim queries using synthetic data generation pipeline.
3. Perform Semantic Search: Leverage LanceDB with sentence embeddings for semantic similarity search to evaluate insurance claim approvals.
4. Route Queries: Implement a Kernel Agent to classify and route queries to either a Loan Agent or an Insurance Agent.

Future enhancements could include adding more sophisticated intent classification or integrating real-time data for loan and insurance predictions.

Setup Instructions

1. Install Dependencies: Ensure you have Python installed. Then, install the required packages:

   pip install pandas joblib pyarrow sentence-transformers lancedb mistralai scikit-learn

2. Prepare Data:

   • The notebook downloads credit_risk_dataset.csv from Google Drive automatically for loan modeling.
   • Insurance claim queries are generated dynamically within the notebook and saved as auto_insurance_claims.csv.

3. Set API Keys:

   • Add your Mistral API key to the environment variable MISTRAL_API_KEY or use the default provided in the code.
   • Add your Hugging Face token as hf_token in the notebook for the SentenceTransformer model (default provided: hf_xxxxxx).

4. Run the Notebook: Open the fintech-ai-agent.ipynb notebook in Jupyter or Google Colab and execute the cells sequentially.

Learn More: Blog

For a detailed explanation of how this system works, check out a hypothetical blog post (you can replace this link with an actual one if you write it):

Read the Blog Post

Google Colab