Find your perfect PhD Advisor with GradRAG

GradRAG streamlines the process of finding the perfect PhD advisor. Whether your interests lie in , , , , , or , GradRAG recommends advisors who align with your research goals—making your search more efficient and personalized.

How the matching works

GradRAG does not search for keywords. It searches for meaning, then asks a language model to explain what it found. Here is the whole process.

See the full architecture
  1. Step 1

    We read your question

    Your question is first checked to make sure it is about research or advisor matching, so an off-topic search fails fast instead of returning noise.

  2. Step 2

    One question becomes many

    A single question rarely matches how researchers describe their own work. We expand it into several differently-worded searches so a good advisor is not missed over vocabulary.

  3. Step 3

    Meaning, not keywords

    Each search is converted into a numeric representation of its meaning. Asking about 'teaching machines to see' can therefore find a computer vision lab that never uses those words.

  4. Step 4

    Searching the advisor index

    Those searches run against an index built from public faculty profiles and publication records, covering research areas, publication counts and university affiliation.

  5. Step 5

    Ranking the candidates

    Results from every search are merged, duplicates removed, and the closest matches kept. An advisor surfacing under several of your searches ranks higher than one found by a single lucky hit.

  6. Step 6

    Explaining the match

    Finally a language model reads the shortlisted profiles and writes each recommendation, saying why that person fits what you asked. Cards appear as they are written, so you can start reading immediately.

What the match score means

Every recommendation carries a percentage. It measures one thing: how closely the meaning of your question resembles the text of that advisor's published work. It is a measure of similarity, not of quality, and not a probability that you would be a good fit together.

Typical score What it usually means
50% and above A clear topical match. The advisor publishes on what you described.
Around 40% Your question was probably too broad to separate one researcher from another. Naming methods or applications will sharpen the results.
Around 30% Little genuine overlap. These are simply the least-distant profiles in the index, not real recommendations.

Why nothing scores 90%

The comparison is between a one-line question and pages of academic writing, so the ceiling is naturally low. Scores cluster in a narrow band, which means a gap of two or three points between two advisors is noise. Treat the top handful as a set to read through, not as a ranked leaderboard.

What it cannot see

The score knows only published text. It does not know whether an advisor is taking students, how they mentor, whether they have funding, or whether they have recently changed field. Those questions are answered by their group's page and by talking to their students.

A note on accuracy. Recommendations and their explanations are written by gpt-5.4-mini, over an index built with text-embedding-3-small. Match scores reflect how closely an advisor's published work resembles your question — they are a starting point for your own reading, not a ranking of quality. Application details such as deadlines, GPA and GRE expectations, and funding are generated by a language model and can be out of date or wrong. Always confirm them on the department's official page before you rely on them.