HuskyGPT
UConn Storrs is home to over 25,000 undergraduates who each have to navigate degree requirements, policies, and information scattered across course catalogs, administrative portals, and Reddit.
The hours students spend hunting down answers add up: students might miss deadlines, incur late fees, and experience unnecessary stress. I built HuskyGPT to fix that.
HuskyGPT is an AI assistant designed specifically for UConn students. Instead of pointing users toward a generic search engine or an LLM trained on the open web, it answers natural-language questions, like:
"How do I switch my major?"
"What are my graduation requirements?"
"What dorms on campus are the best?"
HuskyGPT accomplishes this by drawing from a curated, campus-specific knowledge base. The underlying architecture uses retrieval-augmented generation (RAG), which grounds the model in verified UConn data rather than external sources. This is where more generic tools like ChatGPT fall short: general-purpose models aren't kept current on the niche, institution-specific information students actually need.
The product is designed around two tiers:
A free version handles everyday questions with guardrails to minimize hallucinations
A premium layer, HuskyGPT+, stores conversation history, lets students upload their own documents (syllabi, transcripts), and surfaces personalized recommendations around coursework, clubs, and scheduling. The goal was to replace the friction of student life with something that feels less like a support chatbot and more like a campus mentor.
Beyond UConn, I see this infrastructure being used to scale: the RAG system is white-labelable, meaning other universities could deploy their own version on top of their own data. The core insight is that the AI isn't the moat; the data is.