← Back to home

Temple TU-GO

Smart Campus Navigation System

Role: Lead Developer Type: Capstone Project Focus: AI / RAG

TU-GO is an AI chatbot tailored to the needs of Temple University students, staff, and visitors. It is optimized for university related queries. As the lead developer on my 6-person capstone team, I helped build this AI assistant, and learned a lot about LLMs and how to work with and manage them effectively.

TU-GO landing page screenshot showing the campus navigation interface

Technologies

C# .NET Core MVC Entity Framework AWS PostgreSQL pgvector Google Gemini Tailwind CSS

The Problem

Temple’s information is scattered throughout various sources, TUPortal, Course Bulletins, the main website, a mobile app, and dozens of department sites. A lot of this information is not properly identified by google making it hard to quickly and efficiently access resources.

The Solution

A comprehensive campus navigation hub powered by AI. Users ask natural language questions and get instant answers with clickable source links to official Temple pages.

  • Campus navigation and directions
  • Dining options with descriptions
  • Event discovery
  • Academic resources and deadlines
  • Source citations linking to Temple websites

RAG Architecture

Custom RAG pipeline built in C#/.NET:

12,000+ Pages scraped and indexed
  • Data collection: Scraped 12,000+ pages from Temple's web presence
  • Vector storage: PostgreSQL with pgvector extension
  • Embeddings: Convert queries to vectors for semantic search
  • Generation: Google Gemini produces responses using retrieved context
  • Citations: Every response includes clickable links to source pages
1. Query - User asks a natural language question
2. Embed - Convert question to vector representation
3. Search - Find relevant documents in pgvector
4. Generate - Gemini creates response with source citations

Technical Highlights

  • .NET Core MVC web application
  • Entity Framework ORM
  • AWS Cognito for authentication
  • Admin dashboard for usage analytics and response moderation
  • Tailwind for frontend styling