---
title: Financial Bot
emoji: 🚀
colorFrom: red
colorTo: green
sdk: gradio
sdk_version: 4.16.0
app_file: app.py
pinned: false
license: mit
---

This is the Inference module of a 3-part FTI feature-training-inference RAG-framework LLMOps course. \
In this iteration, I've replaced Falcon 7B Instruct with the currently-SoTa (Jan '24) Mistral-7B-Instruct-v0.2, \
fine-tuned using Unsloth on financial questions and answers generated with the help of GPT-4, quantized \
and augmented with a 4bit QLoRa. \
\
Prompt analysis and model registry is handled by Comet LLM, and finance news is pulled via an Alpaca API, processed \
by Bytewax, and then sent as a vector embedding to Qdrant's serverless vector store. LangChain chains the prompt and \
most relevant news article to provide answers with real-time finance information embedded within the output. \
\
#TODO: Add citations to output to show end-user which article has been used to generate the output.

I have contributed to the original MIT licensed (ka-ching!) course which can be found here:
https://medium.com/decoding-ml/the-llms-kit-build-a-production-ready-real-time-financial-advisor-system-using-streaming-ffdcb2b50714