KatzScholarMind

A Katz-native AI assistant that helps graduate students move from research topic to paper synthesis, faculty matching, events, and next-step outreach in one workflow.

Comment

Inspiration

Research support at Katz is often fragmented. Students may use one tool to find papers, another to summarize them, and manual browsing to identify relevant faculty, events, or program-specific information. Generic AI tools can help with research tasks, but they usually do not understand the Katz ecosystem well enough to guide a student from a topic to a grounded academic next step.

We built KatzScholarMind to solve that problem. Our goal was to create a Katz-native academic assistant that helps students move from topic exploration → paper synthesis → Katz-specific guidance → faculty connection → next-step action in one workflow.

What it does

KatzScholarMind is a research and academic navigation assistant designed for Katz graduate students, especially those working in AI, computer science, cybersecurity, data analytics, and related technical areas.

It helps users:

  • find and organize relevant research papers
  • summarize methods, claims, and findings
  • identify possible research gaps and future directions
  • answer Katz-specific questions using grounded institutional data
  • match a topic to relevant Katz faculty
  • surface relevant Katz or YU events
  • support next-step outreach to faculty
  • generate structured outputs such as citations, study plans, and draft research support

Instead of acting like a generic chatbot, KatzScholarMind connects research exploration to the real Katz academic environment.

How we built it

We built KatzScholarMind as a modular AI platform with two connected layers: a research workflow and a Katz-grounded guidance layer.

The platform includes:

  • a paper discovery and synthesis workflow
  • a Katz-grounded Q&A assistant
  • a paper-focused conversational assistant
  • a curated faculty matching layer
  • an events layer backed by Katz and YU sources
  • a Smart Advisor that connects topic → faculty → event → next step
  • structured outputs such as citations, reading plans, and review-style support

Technically, we used:

  • a FAISS-based retrieval pipeline with persistent disk caching
  • a crawler over Katz and YU website content
  • curated Katz faculty data
  • event fetching with caching and fallback behavior
  • a modular orchestration layer for routing tasks to the right component
  • a working Streamlit prototype for end-to-end interaction

We also added fallback behavior so the application remains usable even when one model path is unavailable.

Challenges we ran into

One of the biggest challenges was balancing research usefulness with institutional accuracy. A general LLM can sound confident, but for Katz-specific questions we needed grounded answers instead of hallucinated ones.

Another challenge was reliability. Website content changes, event pages are noisy, and some institutional pages are not cleanly structured for retrieval. We had to carefully handle crawling, indexing, caching, and fallback behavior so the system stayed practical instead of brittle.

We also had to keep the scope disciplined. Rather than trying to solve every academic workflow, we focused on the strongest use case: helping a Katz student move from a topic to a concrete academic next step.

Accomplishments that we're proud of

We are proud that KatzScholarMind is a working prototype, not just an idea.

Some accomplishments we are especially proud of are:

  • a working research workflow from topic to structured outputs
  • Katz-grounded Q&A using institutional content
  • faculty matching using curated Katz faculty information
  • event matching from Katz and YU sources
  • a Smart Advisor flow that connects research interest to practical next steps
  • citation and study-support features for real academic use

What makes the project different is that it does not just answer questions. It helps connect research support with Katz-specific academic action.

What we learned

We learned that students do not just need answers. They need context, direction, and a next step.

We also learned that grounded academic support is more useful than a long list of loosely connected features. The strongest value of KatzScholarMind is that a student can begin with a topic and end with a clearer starting point, a relevant faculty connection, and a practical action plan.

What's next for KatzScholarMind

Our next steps are to:

  • pilot the tool with Katz students
  • improve output quality and reliability
  • refine event and faculty guidance using real user feedback
  • strengthen paper comparison and synthesis support
  • improve the public demo and deployment experience

In the longer term, we see KatzScholarMind as a model for institution-specific academic AI systems that combine research help with grounded local guidance.

Built With

Share this project:

Updates