KT_Generator

Repository of the code base for KT Generation process that we worked at LifeSight, Google Cloud and Searce GenAI Hackathon.

Repo structure

• Code
• Presentation
• Readme.md

Prereqs

1. ffmpeg
2. poetry

Setup

1. Copy Code/KT Generator/.env.sample to Code/KT Generator/.env and fill in the values.
2. Create a file kt_gen3 in the root directory.
3. Create a folder called test in the root directory and add the codebase you wish to test in it.
4. Create a folder called scripts in the root directory. The scripts generated for the videos will be stored here.
5. Edit the required_nodes with the required functions/classes. a. Class format = class_name b. Function format = class_name.function_name
6. Install deps with $ poetry install
7. Add poe as a poetry plugin $ poetry self add 'poethepoet[poetry_plugin]'
8. Run tests with $ poetry poe test
9. Lint with $ poetry poe lint
10. Run a local build with $ poetry poe all

Note: This is a happy path, so if any errors, report

Run

1. $ poetry poe run

Notes

1. Default model is gpt-4. Pass the required model to service configuration to change.

Code

This folder contains the code base for the KT Generation process.

• KT Generator
  • CodeParser.py: This file contains the code to parse the code base and chunk it into logical code blocks (chunks).
  • CarbonSnippets.py: This file generates the carbon.now snippets for the code blocks.
  • ResponseGenerator.py: This file generates the explaination and the code summary using Llamaindex and PaLM LLM.
  • DIDVideoGenerator.py: This file generates the DID video avatar using the code explainations for all the chunks.
  • CreateVideo.py: This file stitches the final video using the videos and the code snippets and the summary.
  • main.py: This file is the main file that runs the entire process.