Loading projects...
AI-Powered Multilingual E-Learning Backend
30+ Specialized RAG Tutoring Agents
An AI-powered e-learning backend, featuring a multi-agent, multi-modal AI tutor capable of teaching over 30 distinct subjects in multiple languages.
PythonFastAPIMongoDBLangGraphRAGAWS S3AWS ECSJWTOAuth2
Timeline
2 Months
Intensity
2 Contributors
Access
Proprietary
Revision
Mar 2026
The Challenge
Defining the core problem and identified pain points that necessitated this technical intervention.
Educational platforms lack scalable, interactive support. It's difficult to provide 24/7, on-demand, expert-level tutoring for complex subjects, especially in multiple languages, leaving students with static content.
Technical Solution
The architectural and implementation strategy developed to resolve the challenge.
01
Full LMS API
Endpoints for role-based access control (Students, Instructors, Admins), course creation, blog management, student progress tracking, video lectures, auto-graded quizzes, and certificates.
02
Multi-Modal Interface
Users interact via text, file upload, or a full voice-in/voice-out conversation. The voice system uses OpenAI's streaming TTS and Whisper transcription for real-time audio.
03
AI Query-Routing Orchestrator
A central orchestrator receives all new queries and uses an LLM call to classify the user's intent and route it to the correct specialist agent.
04
30+ Specialized RAG Agents
Each agent (e.g., 'MathBot', 'ScienceBot', 'PythonBot', 'PublicSpeakingBot') is a RAG-powered LangGraph chain that connects to its own vectorized knowledge base (trained from a dedicated S3 bucket) for expert, context-aware answers.
Key Contributions
My specific roles, responsibilities, and the technical value I added to the project lifecycle.
Backend Architecture
Architected the entire backend application, including the FastAPI API and MongoDB database schema for courses, users, and progress tracking.
AI Tutor Orchestration
Designed and built the core AI Tutor system, featuring a central LLM-based query classifier and an orchestrator to route requests to 30+ specialized agents.
RAG Agent Template
Developed the standardized RAG agent template used by all bots, integrating chat history, vectorized document retrieval, and dynamic Matplotlib plot generation.
Multi-Modal API Endpoints
Engineered the multi-modal API endpoints for both streaming text (`/invoke-worker`) and streaming voice (`/invoke-voice-worker`).
Voice Conversation System
Integrated OpenAI's Whisper (transcription) and streaming TTS (speech synthesis) to create a seamless, real-time conversational audio experience.
Security & Access Control
Implemented the complete user authentication and security system (JWT/OAuth2) for role-based access control across Students, Instructors, and Admins.
Performance Metrics
Results & Impact
Subject Agents
30+
Deployed 30+ specialized RAG agents covering math, science, programming, public speaking, and more — each with its own vetted knowledge base.
TA Reliance Reduced
90%
AI tutor absorbed 90% of repeat student queries previously handled by human teaching assistants.
Always-On Support
24/7
Students receive instant, expert-level answers in multiple languages at any hour — no scheduling required.
Input Modalities
3
Supports text chat, document/file upload, and full real-time voice-in/voice-out conversation via Whisper and TTS.
Query Routing
< 2s
Central LLM orchestrator classifies intent and routes to the correct specialist agent in under 2 seconds.
User Roles
3
Full role-based access control supporting Students, Instructors, and Admins with distinct permission sets.