Project Overview

Developed an AI-powered chatbot to quickly analyze and interpret 10-K and 10-Q financial documents.

Utilized the XBRL to JSON API to collect 10-K and 10-Q reports for major companies from the U.S. Securities and Exchange Commission's (SEC) EDGAR database: SEC EDGAR Database.

First solutions after processing for clients with a Cloud subscription:

• Built question-answering chatbots using OpenAI's API, Mistral AI API, and Llama3.
• Finetuned medium-sized LLMs like DistilBERT, although training costs are significant.

Production-Ready Solution

Implemented a production-ready solution using text similarity and NER.

Next Steps

• Continue preprocessing the data to enhance model training efficiency and potentially reduce costs.
• Experiment with CopilotKit.
• Utilize quantization techniques to optimize LLMs for use on less powerful machines (e.g., GPUs with 2.6 TFLOPS).