Hackathon
Veluate - AI Teacher Evaluation
2026
About this project
Veluate is an AI-powered teacher evaluation system that links how a lesson was taught to what students got wrong, with video proof. It ingests lecture recordings, syllabi, and student exam papers, then runs a multi-agent LangGraph pipeline to answer one question: where exactly did teaching break down, and how does that map to exam mistakes? Features: 🌡️ Confusion Heatmap: timestamped moments in the lecture most likely to lose students 📐 Structure Report: analyses lesson flow, concept build-up, and narrative coherence 📊 Exam Gap Analysis: extracts weak concept clusters from student exam PDFs 🎯 Cross-Reference Report: links specific teaching moments to exam failure clusters 🎬 Video Clip Evidence: auto-generated timestamped clips of the problematic moments via VideoDB ✅ Fact-Check Agent: verifies lecture claims against live web results through the Bright Data SERP API 🔀 Pluggable LLMs: switch between Anthropic, Kimi, and OpenAI with a single env var ⚡ Live job progress: async FastAPI job API with a real-time Next.js evaluation dashboard
Tech stack
- Python
- LangGraph
- LangChain
- FastAPI
- Next.js
- TypeScript
- Tailwind CSS
- shadcn/ui
- VideoDB
- Bright Data
- Anthropic Claude
- SQLite