PrismAI — Meeting Intelligence

One transcript. Seven AI agents. Instant clarity.

PrismAI transforms any meeting transcript into structured intelligence — summaries, action items, sentiment analysis, follow-up emails, calendar suggestions, and a meeting health score — powered by a multi-agent pipeline on Groq + LLaMA 3.3 70B.

Frontend is ready for deployment on Vercel. The backend remains on Render.

---

How it works

An orchestrator LLM reads your transcript and dynamically routes it to the right specialized agents, all running in parallel:

Agent	Output
Summarizer	Concise 2-3 sentence TL;DR
Action Items	Who owns what, with due dates
Decisions	What was actually agreed or resolved
Sentiment	Tone score + conflict detection
Email Drafter	Ready-to-send follow-up email
Calendar Suggester	Follow-up meeting recommendation
Health Score	0-100 meeting quality score with breakdown

Plus a Chat interface to ask questions about any meeting in natural language.

Input methods

• Paste a transcript directly
• Record live audio via browser microphone (Web Speech API)
• Upload an audio file — transcribed via Groq Whisper large-v3

Stack

Layer	Tech
Frontend	React + Vite + Tailwind CSS
Backend	FastAPI (Python)
AI	Groq API — LLaMA 3.3 70B + Whisper large-v3
Hosting	Vercel (frontend) + Render (backend)

Run locally

# Backend
cp backend/.env.example backend/.env   # add your GROQ_API_KEY
cd backend && pip install -r requirements.txt && uvicorn main:app --reload

# Frontend (new terminal)
cd frontend && npm install && npm run dev

Get a free Groq API key at https://console.groq.com

Deploy frontend to Vercel

1. Import this repo into Vercel.
2. Keep the project root at the repo root so vercel.json is used.
3. Add VITE_API_URL and point it at your Render backend, for example https://meeting-copilot-api.onrender.com.
4. Deploy.

The frontend no longer depends on a GitHub Pages base path, so it works cleanly on a root Vercel domain.