Getting Started

Introduction

PureRecall is an AI-driven meeting transcription and retrieval system designed to handle:

• Real-time Transcriptions: Using AWS Transcribe (audio → text)
• Data Storage & Search: Chunks stored in Supabase, queried through a Vector DB approach (pgvector)
• RAG Summaries: Summaries generated via GPT
• Intuitive UI: React + Tailwind + Vite for frontend, with robust integration via Supabase Edge Functions on the backend

Main Tech Stack

• React (TypeScript) for the frontend
• Supabase (PostgreSQL + pgvector) for chunk storage & indexing, plus Edge Functions for serverless logic
• AWS for real-time transcription using Amazon Transcribe Streaming
• OpenAI (GPT) for summarization, chunk analysis, keyword extraction

Key Motivations

1. Token Cost Management: Instead of sending entire transcripts, we chunk + embed, retrieving only relevant pieces for GPT
2. Speed & Scalability: We use parallel approaches for chunk ingestion (Supabase Edge Functions can be triggered in parallel)
3. Ease of Development: A single codebase with local + production parity (via Supabase CLI)

Installation

1. Clone Repository

git clone [repository-url]
cd [repository-folder]

2. Install Dependencies

npm install

3. Supabase Setup

Install Supabase CLI:

npm install -g supabase
supabase login

For local development:

supabase start
supabase functions serve

4. Environment Variables

Create a .env in the root folder with the following variables:

• VITE_SUPABASE_URL
• VITE_SUPABASE_ANON_KEY
• VITE_AWS_ACCESS_KEY_ID
• VITE_AWS_SECRET_ACCESS_KEY

5. Run Development Server

npm run dev

This serves the React app on a local port (e.g., localhost:5173 or similar).

6. Deploy Edge Functions

When ready to push to production:

supabase functions deploy