EduVoice — Voice-First Learning Assistant

A voice-first AI agent that helps students learn through natural conversation. Built with Vapi (voice AI), Qdrant (vector search), and OpenAI (embeddings + LLM).

Students can call and ask questions about any subject — physics, chemistry, math, biology, history, computer science, and more. The agent retrieves relevant knowledge, generates clear spoken explanations, and maintains conversation context.

Architecture

Student speaks → Vapi (STT) → FastAPI Server
    → Embed query (OpenAI) → Qdrant semantic search
    → Retrieved context + query → LLM (RAG)
    → Response text → Vapi (TTS) → Student hears answer

Tech Stack

• Vapi — Voice interface (speech-to-text, text-to-speech, call management)
• Qdrant — Vector database for semantic knowledge retrieval
• OpenAI — Embeddings (text-embedding-3-small) + LLM (gpt-4o-mini)
• FastAPI — Python backend server
• ngrok — Tunnel local server for Vapi webhooks

Setup

1. Install dependencies

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Configure environment

Copy .env.example to .env and fill in your keys:

cp .env.example .env

You need:

• Vapi API Key — from dashboard.vapi.ai
• Qdrant Cloud URL + API Key — from cloud.qdrant.io (free tier)
• OpenAI API Key — from platform.openai.com

3. Seed the knowledge base

python scripts/seed_knowledge.py

This embeds 25+ educational documents into Qdrant across physics, chemistry, math, biology, history, geography, computer science, and English.

4. Start the server

python -m app.main

5. Expose with ngrok

ngrok http 8000

6. Create the Vapi assistant

python scripts/setup_vapi.py https://your-ngrok-url.ngrok.io

7. Test

Go to dashboard.vapi.ai → find "EduVoice" → click Talk to test via browser.

Features

• RAG-powered answers — retrieves relevant knowledge before answering
• Multi-subject — physics, chemistry, math, biology, history, CS, and more
• Conversation memory — maintains context across turns within a call
• Multilingual — responds in the same language the student speaks (Hindi, English, etc.)
• Voice-optimized — concise, spoken-friendly responses
• Encouraging tone — patient and supportive for learners

Project Structure

Voice-Agent/
├── app/
│   ├── __init__.py
│   ├── config.py          # Environment config
│   ├── main.py            # FastAPI server + Vapi endpoints
│   └── rag.py             # RAG pipeline (Qdrant + OpenAI)
├── scripts/
│   ├── seed_knowledge.py  # Populate Qdrant with knowledge
│   ├── setup_vapi.py      # Create Vapi assistant
│   └── test_call.py       # List/test assistants
├── data/
│   └── knowledge.json     # Educational knowledge base
├── requirements.txt
├── .env.example
└── README.md

Hackathon

Built for the Voice AI Hackathon — Track 3: Accessibility & Societal Impact (Education).