UET GPT - Your AI Guide to UET Taxila
An intelligent AI chatbot that answers any question about the University of Engineering and Technology (UET) Taxila - admissions, fee structure, academic programs, departments, faculty, campus life, transport, hostels, scholarships, and more. Powered by RAG over official UET Taxila data.
One-line description (SEO): UET GPT is an AI chatbot that gives instant, cited answers about UET Taxila - admissions, fees, programs, faculty, and campus life - using RAG over the university's official data. Licensed under AGPL-3.0.
UET GPT is an autonomous RAG (Retrieval-Augmented Generation) chatbot for UET Taxila. It
answers questions about admissions, departments, fees, exams, faculty, and
campus life using a hybrid search + LLM generation pipeline over a corpus of
crawled university web pages and ingested PDFs.
Live app: https://uet-gpt.vercel.app · Press kit
What is UET GPT?
UET GPT is a specialized AI assistant for the University of Engineering and Technology (UET), Taxila - one of Pakistan's premier engineering institutions, founded as a UET Lahore campus in 1975 and granted its independent charter in 1993, now serving 5,000+ students across 30+ programs. Instead of digging through scattered university web pages and PDFs, students, applicants, parents, and faculty ask UET GPT in plain language and get fast, accurate answers backed by citations from official UET Taxila sources.
Key things to know:
- Built for UET Taxila - every answer is grounded in UET Taxila's official website (
web.uettaxila.edu.pk), admissions portal (admissions.uettaxila.edu.pk), and ingested documents.
- Retrieval-Augmented Generation (RAG) - answers are retrieved from a continuously crawled, indexed knowledge base, not hallucinated from generic model memory.
- Cited and verifiable - responses link back to the official university material they came from.
- Open source - the full stack is released under AGPL-3.0 and is free to use.
UET GPT is currently free to use at https://uet-gpt.vercel.app.
Stack
- Frontend: Next.js 16 (App Router, Turbopack), React 19, Tailwind CSS 4, Radix UI
- Backend: Convex (real-time DB + serverless functions, native vector index)
- Auth: Clerk (RBAC: user / admin / superadmin)
- LLM: Vercel AI SDK with a Groq → Cerebras → Gemini fallback chain
- Crawler: Python async BFS crawler (
curl_cffi, trafilatura)
See architecture.md - the single source of truth - for the
full system design.
Prerequisites
- pnpm (this repo is pnpm-locked; do not mix npm/yarn - see
architecture.md §17)
- Node.js (version per
package.json / .nvmrc)
- Python 3.x (for the async BFS crawler and PDF ingest scripts in
scripts/)
- A Convex account, a Clerk application, and API keys for Groq / Gemini / Cerebras
Environment Variables
Set these in .env.local (Next.js) and in the Convex dashboard as appropriate.
See .env.local.example for the full list with comments, and architecture.md §8.2 for the authoritative descriptions.
Getting Started
pnpm install
npx convex dev
pnpm dev
Open http://localhost:3000 to use the app.
Testing
pnpm test
pnpm lint
pnpm test:e2e
Documentation
architecture.md - single source of truth for system designAGENTS.md - agent / contributor instructionstesting.md - testing strategy (trimmed, agent-optimized)THREAT_MODEL.md - security threat modelCRONJOB.md - autonomous maintenance protocolDESIGN.md - design system (Neo Kinpaku)PRODUCT.md - product definitionCHANGELOG.md - engineering work logdocs/ - focused reference docs:
docs/reference/ - API signatures, env vars, auth matrix, component inventorydocs/security.md - security controls summarydocs/embedding-strategy.md - embedding model + vector storedocs/agent-coordination.md - frontend/backend agent boundary rulesdocs/frontend_backend_boundaries.md - ownership rules between src/ and convex/docs/deployment.md - deployment architecturedocs/crawl-robots-audit.md - robots.txt compliance decisions