Supervisor Outreach Agent

Project Story — Supervisor Outreach Agent

About the Project

Finding a graduate research supervisor is often fragmented and time-consuming. Students must compare universities, search faculty directories, read research pages, identify relevant publications, and determine whether professors may be accepting students.

Supervisor Outreach Agent makes that first stage easier. A student enters their target country, region, degree level, and research field. The system then discovers relevant universities, finds potential supervisors, builds research profiles, and ranks candidates with clear explanations.

Inspiration

The project was inspired by the difficulty students face when searching for Master's or PhD supervisors. Search engines return scattered pages, faculty websites vary widely, and important information is often buried in university directories or research-lab pages.

We wanted to build a tool that transforms this research process into a guided workflow while keeping the student in control of the final decision.

How We Built It

The project uses a Next.js frontend and a FastAPI backend.

The frontend provides:

• A validated search form for country, region, degree, and research field.
• A live loading dashboard that shows progress for each agent stage.
• Ranked supervisor cards and detailed professor pages.

The backend runs a four-stage AI-assisted pipeline:

1. University Finder : discovers relevant universities.
2. Faculty Finder : searches for researchers related to the selected field.
3. Research Profiler : extracts research summaries and recent work.
4. Supervisor Ranker : evaluates candidates and produces a fit score with reasoning.

Challenges We Faced

The biggest challenge was that university websites are inconsistent. Faculty information may appear on department pages, personal sites, lab pages, PDFs, or incomplete directories.

We addressed this by adding defensive handling for:

• Broken or relative profile URLs.
• Duplicate researchers.
• Social media, directory, and aggregator links.
• Scraping failures and unreadable page content.
• PDF-based research material.
• Missing “accepting students” information.

Another challenge was preventing multiple users from overwriting shared result files during simultaneous searches. We added a backend lock so that only one search pipeline can run at a time.

What We Learned

This project taught us that AI systems are most useful when they are transparent. Rather than returning unexplained recommendations, the application provides fit scores, research summaries, recent work, and reasoning for every ranked candidate.

We also learned that real-world AI applications need more than an LLM call. Reliable results require validation, search fallback strategies, scraping safeguards, structured output parsing, error handling, and a user interface that communicates progress clearly.

Collaboration with Codex and GPT-5.6

Codex accelerated the development workflow by helping connect the frontend and backend, refine the multi-agent pipeline, improve validation, handle edge cases in scraped data, and iterate quickly on the user experience.

GPT-5.6 contributed reasoning and implementation support throughout the project: translating requirements into components, identifying failure cases, improving the data-processing workflow, and helping document the final architecture.

The key product decisions remained human-led: focusing on supervisor discovery rather than automated outreach, prioritizing transparency in rankings, using live progress updates, and ensuring that students can verify every recommendation before contacting a professor.

Built With

• api
• asgi
• asyncio
• css
• fastapi
• gpt
• next.js
• openai
• python
• react
• tailwind
• tailwind-css-|-|-backend-|-python
• typescript
• uvicorn