Role
Full-Stack Lead
Year
2026
Tech Stack
7 Technologies
Status
Completed
Overview
AI-generated educational content is shallow — most tools summarize Wikipedia and call it a course. TheTutor takes it seriously. A learner types a topic and receives a complete multi-module course grounded in real published textbooks, not hallucinated content. A custom MCP server discovers relevant books, parses them, and feeds structured source material to the AI before any content is generated. The platform also includes voice narration, an in-lesson AI assistant, quizzes, certificates, and a community marketplace. 5-person team, 3 weeks — I served as full-stack lead, contributing 93 of 189 commits and roughly half the codebase.
The Problem
Online learning is fragmented. Generic AI tools produce shallow, hallucinated content with no grounding in real academic material. Learners who want to study a new subject end up cobbling together YouTube playlists, blog posts, and random PDFs — with no structure, no assessments, and no way to verify what they're learning is accurate. There was no platform that could take a single topic and produce a complete, multi-module course grounded in real published textbooks.
Screenshots
Challenges
Orchestrating a four-phase generation pipeline — book discovery, outline generation, parallel lesson generation, and video enrichment — that broadcasts real-time progress, survives server restarts, and handles rate limits without silently dropping lessons.
Making AI-generated lessons feel academically rigorous, not like generic summaries. Different subjects require fundamentally different structures, and content depth needed to scale with the learner's level.
Building voice narration with synchronized text highlighting that works across different browsers and degrades gracefully when services are unavailable.
Solutions
Built a job-based pipeline where lessons generate in parallel but write atomically to individual slots. Failed lessons get a second pass, and orphaned jobs resume automatically on server restart.
Built genre-aware content generation that detects the course subject and adapts the lesson structure accordingly. Citations from discovered textbooks are injected so the AI grounds content in real sources.
Built a layered voice system with neural TTS for narration, synchronized text highlighting, playback speed control, and voice input for the AI assistant — each feature degrades silently when unavailable.
Key Highlights
MCP-Grounded Content: Course content is sourced from real published textbooks discovered and parsed by a custom MCP server — not hallucinated from training data.
Crash-Resilient Generation Pipeline: Four-phase pipeline with parallel lesson generation, automatic crash recovery, and real-time progress streaming to the client.
Voice Narration & AI Assistant: Neural TTS with synchronized text highlighting, playback speed control, and an in-lesson AI assistant that accepts voice input.
Open Source Team Project: 5 contributors, 189 commits, ~22,400 lines of TypeScript. MIT licensed with a published video demo.
Scale & Scope
~22,376
Total Lines of Code
71 components, 24 services
Source Files
22 pages, ~27 endpoints
Pages / Endpoints
8
Database Models
189 (93 mine)
Git Commits
Technology Stack
Outcome
Live at thetutor.vercel.app and open source on GitHub. A 5-person team shipped ~22,400 lines of TypeScript in about 3 weeks, with a published video demo.
