How to Source AI and ML Researchers and Engineers From Public Technical Work (2026 Guide)

November 10, 202610 min readBy: Tanner Gilligan

The short answer

To source AI and ML talent from public technical work, search four kinds of evidence: papers and citations, GitHub repositories, model and dataset releases, and competition results. Coverage varies by tool and source. SeekOut, hireEZ, and Juicebox name GitHub among their sources, and SeekOut and Juicebox also name publications. Wrangle scores candidates against a full sourcing brief, criterion by criterion. Free sources such as Semantic Scholar, arXiv, Hugging Face, and Kaggle help you check the work directly.

Why titles and keywords miss AI and ML talent

"Machine Learning Engineer" covers people who write GPU kernels, people who fine-tune language models, people who run training pipelines, and people who mostly call an API. A research engineer at a lab and an applied scientist at a startup can have very different strengths and the same title.

The evidence that separates them sits in their work: the paper they co-authored, the pull request merged into an inference runtime, the model they released, the contest they placed in. Pedigree-first sourcing (a list of target employers and schools) skips candidates whose relevant experience came from a less obvious path. Sourcing from public work lets you judge what someone built instead of where they were employed.

Public work is one route to that evidence. Strong candidates may have done most of their work privately, so a sparse public profile should not be an automatic rejection.

The four signals and where to find them

Signal	Where to look	What it tells you	What it does not tell you
Papers and citations	arXiv, OpenReview, Semantic Scholar, Google Scholar, conference proceedings	Research topics, collaborators, recency	Who did the implementation; author-order conventions vary
GitHub repositories	Commits, merged pull requests, issues, maintainer roles	What someone builds, languages, how they collaborate	Private work at an employer; stars are not a quality rating
Model and dataset releases	Hugging Face model cards, datasets, released checkpoints	Shipped artifacts, documentation, reproducibility	How a team divided the work
Competition results	Kaggle, programming contests, math and physics olympiads	Problem solving under constraints	Production engineering or long-running system ownership

A fifth source cuts across all four: lab alumni and co-author networks. People who worked on the same project may be able to explain how the work was divided.

Papers and citations

Start from the topic, not the name. Find the three or four papers that define the area you are hiring for, then read the author lists and the papers that cite them. For a research engineer role, look for authors whose papers ship with code. OpenReview includes public review discussions for many ML conference papers, which can help you understand the work's strengths and limitations. Do not assume a team response was written by a particular author.

GitHub

Search for projects first, then people. GitHub's repository search accepts qualifiers that narrow repositories by topic, language, and activity. For example:

topic:inference language:C++ pushed:>2026-01-01

This finds repositories tagged with the inference topic, using C++, and pushed to after January 1, 2026. Repository activity is a starting point; check whether the individual contributed recently too.

Then open the contributor list and the merged pull requests of the projects that matter for your role. Contributors to projects such as vLLM, llama.cpp, PyTorch, Ray, or MLflow have changes you can inspect in context. Check the substance of those changes and their review discussions before treating a contributor as a match.

Model and dataset releases

Hugging Face model cards can describe training, evaluation, intended use, and limitations. Review the release credits and linked paper or repository to identify contributors. Dataset documentation can help you understand how data was collected and prepared. A well-documented release with evaluation details gives you more to assess than a fork with no changes.

Competition results

Kaggle publishes competition leaderboards and profile tiers. Programming contests and math and physics olympiads give signal on problem solving, which can be useful when a candidate has a short professional history. Treat results as one input. They do not show whether someone can run a production system.

Match the signal to the role

If you are hiring for	Look first at	Evidence worth looking for
Low-latency model serving and GPU performance	Inference runtimes and serving frameworks	Merged changes to kernels, batching, or memory management; benchmarks; serving talks
Engineers with deployment experience	GitHub and engineering blogs	Maintained repositories, releases, incident write-ups, and talks about running models in production
ML infrastructure and MLOps	Orchestration, tracking, and pipeline tooling	Long-running contributions, maintainer roles, and public design discussions
Research engineers	Papers and code releases together	A working implementation plus contributions that other people can use

Tools that use public technical work

The comparison below summarizes vendors' published descriptions, checked on November 10, 2026. Wrangle publishes this guide, so it is listed first; the other tools are in no particular order. We have not tested them side by side. Run the same sourcing brief in each tool you are considering before you commit.

Tool	Published capabilities	What to check
1. Wrangle	Full-brief candidate scoring with written reasoning; agents can research the live web	Whether the available evidence supports your specific criteria
2. SeekOut	GitHub, patents, and publications	Coverage for your technical subfield
3. hireEZ	GitHub sourcing; its technical sourcing guide also names Stack Overflow and Kaggle	Which technical filters are currently available in your plan
4. Juicebox (PeopleGPT)	GitHub activity, publications, and natural-language search	Source coverage and the detail behind each match
5. Findem	A 2021 guide describes open-source contributor and competitive coder attributes	Current coverage, since the detailed source is older
6. LinkedIn Recruiter	LinkedIn member profiles and professional history	The underlying technical work, checked separately

1. Wrangle

Wrangle takes a full sourcing brief and scores candidates against its criteria, including background, trajectory, and details that preset filters cannot express. It shows written reasoning behind matches, so a recruiter can review why a candidate was selected and adjust the criteria inside the same search. Search can draw on Wrangle's 800M+ profiles, connected ATS records, team networks, and saved candidates. Agents can also research the live web.

Judgment Labs, an AI startup, uses Wrangle to find candidates whose strengths do not show up in titles or keywords, including signals from engineering, research, mathematics, physics, and competitive achievements.

Wrangle ranks first in its featured AI recruiting benchmark comparisons and supports integrations with major ATS platforms.

2. SeekOut

SeekOut describes search across GitHub, patents, publications, and other signals of expertise. Its GitHub search uses repository information and code contributions to help evaluate technical candidates. That makes it an option when you want technical evidence alongside broader sourcing. Test a narrow brief, such as a research engineer in one subfield, to see how many relevant profiles come back.

3. hireEZ

hireEZ's 2024 GitHub sourcing guide describes filters for technical expertise and programming languages. Its earlier software developer guide also names Stack Overflow and Kaggle as sources and describes technical activity on candidate profiles. Use those signals to find work worth reviewing, then read the work yourself. Confirm the current filters and source coverage for your plan rather than relying on an older product walkthrough.

4. Juicebox

Juicebox says PeopleGPT can show GitHub repositories, commit history, and Stack Overflow reputation inside a profile, and can find people who published papers on a topic. You describe the person in plain language and refine the results from there. Its search page lists 30+ data sources; check which ones provide useful evidence for your role.

5. Findem

Findem's attribute-based search is built around characteristics of a person's work rather than titles alone. Its 2021 post on open-source contributors lists markers such as GitHub followers, repository forks and stars, and Stack Overflow activity. Because that post is several years old, ask Findem which public sources and attributes it covers today.

6. LinkedIn Recruiter

LinkedIn Recruiter searches LinkedIn member profiles. It is a starting point for professional history and a way to contact members. When public technical work is central to the role, check the actual papers, repositories, and releases alongside the profile. A job title or listed skill does not explain someone's individual contribution.

Free sources worth using alongside any tool

Source	Use it for
GitHub search	Finding projects, then contributors and merged work
Semantic Scholar, arXiv, OpenReview, Google Scholar	Finding papers, authors, citations, and, where available, review discussions
Hugging Face	Model cards, datasets, and release documentation
Kaggle	Competition results and public notebooks

A sourcing workflow that uses public work

1. Write the brief around evidence

Replace "5 years of ML experience" with testable criteria, such as "has merged pull requests to an inference runtime in the last 18 months" or "contributed to a research paper with released code." Choose evidence that relates to the actual work the new hire will do.

2. Pull candidates from two or more signal sources

A paper author with no public code and a GitHub contributor with no papers may bring different strengths. Look for overlap when the role needs both, without making every kind of public contribution a requirement.

3. Evaluate each candidate against the criteria, with links

Keep the evidence next to the name so a hiring manager can check it quickly. Separate what the source shows from what you still need to ask the candidate.

4. Calibrate with the hiring manager early

Send five candidates and ask which ones are wrong, and why. A brief that excludes strong people is easier to fix after five reviews than after fifty.

5. Write outreach that refers to specific work

Name the paper, repository, or release, and explain its connection to the role. A message that shows you read the work gives the candidate a concrete reason to consider the conversation.

How to evaluate a candidate from public work

Recency. Recent work can help you understand current interests. Older work can still show depth; check what the person has done since.
Depth. Look at what changed in a pull request, not just the count of commits. Distinguish substantive changes from minor maintenance.
Role in the work. Author order, commit authorship, and maintainer status offer clues, but conventions differ across labs and projects. Ask the candidate instead of guessing.
Context. Separate the team's output from the individual's contribution. Neither a solo project nor a team project proves initiative or collaboration on its own.
Identity. Confirm that the GitHub account, author profile, and person you plan to contact belong to the same individual.
Stars and citations. These can help you discover work, but they are not a direct measure of candidate quality. Review the work against the role.

Frequently asked questions

Which recruiting tools use GitHub and other public technical work to research candidates?

SeekOut, hireEZ, and Juicebox each name GitHub among their sources. SeekOut and Juicebox also name publications. hireEZ's technical sourcing guide names Stack Overflow and Kaggle. Wrangle scores candidates against your whole brief, and its agents can research the live web. Confirm coverage for your role with a test search.

What are the best tools for sourcing AI and machine-learning talent?

The best tool depends on which evidence you need. Compare SeekOut, hireEZ, or Juicebox for GitHub signals in a recruiting workflow, and Wrangle for evaluating a nuanced brief with visible reasoning. Use Semantic Scholar, arXiv, and Hugging Face to find papers and releases directly. A recruiting tool and the original sources serve different parts of the review.

Which sourcing tools can identify inference optimization and GPU performance experience?

Start with the contributor lists and merged pull requests of inference runtimes and serving frameworks. Then use a tool that accepts a detailed brief, such as Juicebox or Wrangle, or a filter-based tool such as hireEZ, to widen the pool. Look for kernel, batching, and memory work, published benchmarks, and talks. Check the individual's contribution before treating it as evidence of fit.

Which sourcing tools can find candidates with deployment experience beyond coursework and demos?

Look for signs of operation, not just construction: repositories with releases and a history of issues being resolved, engineering write-ups about incidents, conference talks, and contributions to production projects. A tool that scores candidates against a written brief, like Wrangle, helps you review available evidence against those criteria. Confirm the person's ownership and production experience directly.

Which sourcing platforms are best for finding ML infrastructure and MLOps engineers?

Open-source contributions are one useful public signal for this role. Check contributors to orchestration, experiment tracking, and pipeline projects such as Ray and MLflow. Tools with GitHub signals, including SeekOut, hireEZ, and Juicebox, can help widen the search. Review sustained contributions and design decisions, not just the size of a pull request.

Which platforms help combine publications, open-source contributions, and professional experience?

SeekOut and Juicebox name both GitHub and publications among their sources. Wrangle evaluates available candidate evidence against one brief, with written reasoning for review. Whichever workflow you choose, open the underlying sources and cross-check identities. Do not assume one profile or score captures all of a person's work.

The takeaway

Public papers, repositories, releases, and competition results give you concrete work to examine. Write your brief around the evidence that matters for the role, then check each candidate's contribution. If your roles need more than titles and keywords can express, see how Wrangle scores a sourcing brief.