soutot/upload-ai

Simple project using nlw-ai-mastery by @rocketseat as base.

It's a fullstack app using NextJS 13.4, Postgresql, and Prisma to give users the ability to upload a .mp4 video, add topics, add/select/edit a prompt template, select temperature, and send a request to OpenAI API to get back a response based on the video transcription + sent prompt. In this project we have the use case of a user requesting OpenAI to generate a title and a description for an YouTube video based on the uploaded video content.

This project uses:

• WebAssembly (FFMPEG) to convert the video to audio
• OpenAI whisper to transcript the audio to text
• OpenAI chat to generate suggestions
• ShadcnUI to create UI elements
• Docker to manage the environment containers
• pnpm to manage modules

Getting Started

First, run the development server:

Create a .env file

touch .env

Create a tmp folder under public directory:

cd public && mkdir tmp

Edit it with your data as the following

# DATABASE
DATABASE_URL=""
POSTGRES_USER=
POSTGRES_PASSWORD=
POSTGRES_DB=

# OPENAI
OPENAI_API_KEY=

Install dependencies

pnpm install

Run docker compose

docker-compose up --build

If everything works correctly, the app should be available at http://localhost:3333

Known issues / TODOs

• Currently, prisma seed is not working properly
• Clean unused/unnecessary code
• We could use server actions
• Add more validations and UI improvements
• Use langchain
• Implement Llama2