Automatic Knowledge Transfer (KT) Generation for Code Bases
ravi theja (~ravi1) |
Description:
Introduction: In today's rapidly evolving IT and Software Development landscapes, knowledge transfer (KT) is a pressing challenge. Fragmented knowledge, varied tech stacks, and quick project turnovers can hinder efficient KT processes.
Objective: Streamline the KT process, making it more comprehensible and accessible by auto-generating video explanations paired with code snippets using LlamaIndex.
Methodology:
Code Parsing:
- Use Python’s ast library to segment the codebase into individual blocks.
Summary and Explanation Generation with LlamaIndex:
Utilize LlamaIndex’s ListIndex for an overarching code summary.
Provide context-rich explanations for each code block.
Video Creation with D-ID:
Generate avatar-driven videos to articulate code snippets.
Incorporate voice using Microsoft’s text-to-speech synthesizer.
Video-Code Integration:
Enhance code snippets visually with the carbon library.
Integrate using the moviepy library for a holistic viewing experience.
End Vision: Imagine a platform, potentially named KodeTube(KT), where an organization's entire codebase is elucidated through engaging videos, making KT both seamless and effective.
Prerequisites:
Participants should have a fundamental knowledge of Large Language Models such as ChatGPT.
Speaker Info:
Speaker - 1. Ravi Theja is an Open Source Contributor at LlamaIndex and Senior MLE at Glance (Inmobi).
- Recognised as Top 5 GenAI Experts in India by Analytics Vidhya at Data Hack Summit.
- Contributed different data loaders and evaluation modules to LlamaIndex.
- Built Recommender Systems, NLP, and GenAI applications at Glance. I am part of Glance TV team which is a product built entirely using LLM's and vector DB's. Demo
- He has published papers at ACL and COLING workshops.
- He had given talks about LlamaIndex at the Hasgeek GenAI meetup, Fifth Elephant Conference, Analytics Vidhya Data Hack Summit, and different VC firms (Accel, Together, Speciale).
Speaker - 2:
Vibhav is an MLE-2 at Glance India. He is currently working on recommendation systems and edge computing at Glance from past two years. Before this, he was deep into academic research with multiple publications in NLP, Information Retrieval etc in A* conferences like IJCAI, EMNLP, IEEE BigData. He has been working with Large Language Models (LLMs) even before the ChatGPT era began.