UrGPT
Empowering Conversations, Unleashing Knowledge: UrGPT - Your Intelligent Companion in Code, Documents, Web and more!
Created on 4th February 2024
•
UrGPT
Empowering Conversations, Unleashing Knowledge: UrGPT - Your Intelligent Companion in Code, Documents, Web and more!
The problem UrGPT solves
In the realm of Generative AI applications like ChatGPT, reliance on free resources presents challenges, including privacy concerns with sensitive data and functional limitations affecting reliability. Our application, employing open-source Large Language Models (LLMs), aims to enhance ChatGPT's accuracy and speed. By seamlessly integrating PDF and web page-based answering features as external knowledge sources, this approach seeks to improve performance, address privacy concerns, overcome functional limitations and reducing model hallucination. Utilizing open-source LLMs underscores our commitment to refining Gen Ai for more effective real-world applications.
Challenges we ran into
Q - Running LLM’s on Local Machine
A - Quantizing model to efficiently run on Local Machines, with Googles
free resources available.
Q - Large Response Time to queries
A - Implementing Streaming responses.
Q - Dealing with larger documents, with small context window
A - Using Vector Stores, to retrieve only relevant documents.
Q - Maintaining Chat Context
A - Using SQL database to store previous chats, and provide them as
chat history to the model to continue the previous conversion.
Tracks Applied (1)
Ethereum + Polygon Track
Polygon