How matching works.

From your resume
to the right fits.

Here is how Joblook decides which roles reach your For You list, step by step, in plain words. Every step names the tool or research behind it.

Find my matchesSee the research
Job cards walk a path through a sieve, past a magnifying glass and a clipboard, and a few step through an open door.

The short version.

Five steps from a long list
to a short one.

This runs in the background each day for your primary resume. It never runs while you wait on a page.

  1. Your limits first

    Only roles that fit the role, place, experience and skills you chose.

  2. A wide net

    Keyword search plus the newest roles, up to 250 to start.

  3. Meaning, not words

    We rank by what the work means and keep the best 40.

  4. Every requirement

    Each one is checked against your resume, one at a time.

  5. A plain label

    A fixed rule grades it, and we confirm the role is open.

Steps 1 and 2.

Start with what you asked for,
then cast a wide net.

We only look at open roles that fit the role, location, experience range and skills you set, and we leave out jobs you've applied to or hidden. From those, keyword search finds up to 200 roles that match your resume and preferences, and we add the 50 newest so fresh postings aren't missed.

  • Your preferences are firm limits, not hints
  • Experience ranges overlap: a 2–6 year preference fits roles asking for 1–4 or 5–8
  • Jobs you applied to or marked not interested stay out

Under the hood. Keyword search runs on PostgreSQL full-text search [4].

Step 3.

Compare meaning,
not just keywords.

People describe the same work in different words. We turn your resume, your desired roles and each candidate role into an embedding, a list of numbers that captures meaning, so similar work lands close together. Then we blend three rankings, by meaning, by keywords and by how new the role is, and add how many of your chosen skills it asks for. The top 40 move on.

Same work, different words

Frontend engineeris close toReact developer

Customer successis close toAccount management

People operationsis close toHR generalist

Under the hood. Embeddings come from OpenAI's text-embedding-3-small [3]. The three rankings are blended with reciprocal rank fusion [2]. Research on resume and job embeddings [5] [6] shows why meaning beats keywords alone.

Step 4.

Check every requirement,
one at a time.

When a role is posted, we pull out up to 24 requirements, mark each as required or preferred, keep any "or" alternatives, and save the exact line it came from. Every quote is checked against the original posting. Then Jev, a decision model, reads your resume and gives each requirement one of four answers.

Product Designer, Design Systems

Figma · New York, NY · Hybrid · Quoted from the posting

  • Required

    “5+ years of product design experience”

    Meets

  • Required

    “Experience building or maintaining a design system”

    Meets

  • Preferred

    “Comfortable reading front-end code”

    Partial

  • Preferred

    “Bachelor's degree in design or a related field”

    Unknown

  • Result

    Good match
An example of one role's checks. Your For You list shows the label; the per-item answers stay behind the scenes.

Under the hood. Requirement checks use Jev by TypeSafe [1], which answers with a fixed set of choices instead of writing text. The four answers are meets, partial, doesn't meet and unknown.

Step 5.

Grade it plainly,
then make sure it's open.

A fixed rule, written in our code rather than left to a model, turns those answers into a label.

  • Good matchEvery required item is met. Preferred items never hold a role back.
  • Moderate matchNothing required is clearly missing, but something is partial or not on your resume.
  • Left outA required item is clearly not met, so the role doesn't reach your list.

Before a role reaches you, we confirm it's still open, from the company's own job feed or a fresh check of the posting. That check counts for 12 hours. Your list keeps every Good match and fills up to 20 places with the best Moderate matches.

Under the hood. Open roles are confirmed through applicant tracking feeds such as Greenhouse, Lever, Ashby and BambooHR, or a direct check of the job page.

What this means for you.

Clear rules,
no black boxes.

  • Required means required

    A role that clearly misses a must-have never reaches your list.

  • Unclear isn't a no

    If your resume doesn't say, the role is labeled a Moderate match instead of being ruled out.

  • Nice-to-haves never block

    Preferred items can't remove a role from your list.

  • Labels aren't predictions

    They compare your resume with the role. They don't predict an interview or offer.

  • You decide where to apply

    Joblook never applies for you. You apply on the employer's site.

  • Your resume, your call

    It's used only to match you with roles, and you can delete it from Resumes.

Research and sources.

The work
we build on.

We name the tools we use and the research that shapes our thinking. The results in these papers belong to their authors; they aren't claims about Joblook.

Used in Joblook

[1]Model

Introducing System One Models & Jev

TypeSafe AI · 2026

Checks each requirement against your resume and answers from a fixed set of choices.

Read the source
[2]Paper

Reciprocal Rank Fusion Outperforms Condorcet and Individual Rank Learning Methods

Gordon V. Cormack, Charles L. A. Clarke, Stefan Büttcher · SIGIR 2009

The method we use to blend the meaning, keyword and recency rankings into one.

Read the paper
[3]Model

New embedding models and API updates

OpenAI · 2024

text-embedding-3-small turns your resume and each role into embeddings we can compare.

Read the source
[4]Docs

Full Text Search

PostgreSQL documentation

The keyword search that gathers the first set of candidate roles.

Read the docs

Research that shapes our approach

Joblook doesn't run these models. They study the same problem, matching resumes to jobs, and inform how we think about it.

[5]Paper

ConFit: Improving Resume-Job Matching using Data Augmentation and Contrastive Learning

Xiao Yu, Jinzhong Zhang, Zhou Yu · ACM RecSys 2024

Shows that learned embeddings of resumes and jobs can rank matches better than keyword search alone.

Read the paper
[6]Paper

ConFit v2: Improving Resume-Job Matching using Hypothetical Resume Embedding and Runner-Up Hard-Negative Mining

Xiao Yu, Ruize Xu, Chengyuan Xue, Jinzhong Zhang, Xu Ma, Zhou Yu · Findings of ACL 2025

Builds on ConFit with model-written example resumes and harder training examples. A reference point as we learn from your feedback.

Read the paper

Ready to see
your matches?

Sign in, add your resume, and set your preferences. Your first matches arrive in For You.

Find my matchesBrowse jobs

Free during the MVP launch