dagiteferi/teleexam-ai-vcall-

AI-powered real-time video call tutor microservice for Ethiopian university students preparing for CS exit exams. Built with LangGraph, LiveKit, pgvector, and hexagonal architecture. # TeleExam AI — Video Call Tutor A real-time AI tutoring backend for Ethiopian computer science exit-exam preparation. The project combines FastAPI, WebSockets, Postgres with pgvector, Redis caching, LangGraph orchestration, and LiveKit media sessions to deliver a tutoring experience where each student turn is classified, enriched with curriculum retrieval, and answered with an AI tutor response. > [!NOTE] > This repository contains the backend service and AI orchestration layer. It does not include a frontend client or mobile app. --- ## Badges -blue?style=for-the-badge) --- ## Table of Contents - Project Title - Badges - Features - Architecture - Tech Stack - Project Structure - Installation - Environment Variables - Running the Project - API Documentation - Database - AI Components - Configuration - Security - Testing - Deployment - Development Workflow - Performance Considerations - Troubleshooting - Roadmap - Contributing - License - Acknowledgements --- ## Features - Real-time tutoring session lifecycle with Postgres-backed session tracking - FastAPI HTTP API for session start/end and learner profiles - WebSocket-based live interaction at `/vcall/ws/{session_id}` - LangGraph-based intent routing for: - concept search - YouTube lookup - YouTube summary - exam-question RAG - memory/profile lookup - general tutoring - Redis caching for intent classification, conversation state, RAG hits, and learner profiles - Curriculum ingestion pipeline that reads local text materials and stores embeddings in Postgres via pgvector - Vector similarity search over curriculum chunks using cosine distance - Learner profiling with weak topics, average score, and exam count - Built-in admin ingestion endpoint for curriculum loading - LiveKit room creation and participant token issuance - Extensible provider architecture via ports and adapters for LLM, STT, TTS, embeddings, search, and media --- ## Architecture This project follows a layered, adapter-based archit …