Investment Evaluation by Funds Using LLM

This project, developed using Google Colab, aims to create a Large Language Model (LLM) to evaluate wheter an investment fund can invest in a specific type of asset according to current regulations.

Project Description

The goal of the project is to create a system that uses LLM to answer questions related to investment regulations. The system will be able to determine if a fund can invest in a specific asset, taking into account Brazilian laws and regulations.

Project Structure

1. Database Construction

   • Question and Answer Database: Create a database that coains both easy and difficult questions, along with the answers and the process of understand each answer.

2. Prompt Writing

   • Define and write the prompt for the model.

3. Law Importation

   • Import the brazilian regulation for funds (CVM 175) and fund prospectus

4. Model Testing Models to Test:

   • Native GPT: Comparison between different models.
   • Native GPT + prompt training with the modeling database.
   • Agents.
   • RAG (Query).
   • Agents + Memory (chat) + prompt training.

5. Code

   The code developed in Google Colab is divided into the following sections:

a. Law and Regulation Importation

• Code to import and process the necessary laws and regulations.

b. Native GPT -Refinement of the prompt for the native GPT model.

c. Agents

• Agent implementation

d. Tool for Importing Fund Regulations

• Create a tool for importing fund regulations.

e. Prompt Refinement

• Adjustments and improvements to the prompt for different test scenarios.

f. Test Importation

• Importing test data to a .xlsx file.

6. Test Database

A test table will be used with the following columns:

• Fund Name
• Asset Name
• Allowed (whether the fund can invest or not)
• Difficulty (1 to 5)

7. Requirements

Google Colab: To run the code and perform the tests. GPT Model: Access to the GPT model to run the tests. .xlsx File: File containing the test database.

Contact For more information or questions, contact via pfgom.ita@gmail.com or gigi.bduarte@gmail.com