Plain-language reference for how Halo classifies faculty researchers and how their pages appear on halo.science.

The short version. Halo aggregates publicly available scholarly data — publications from OpenAlex, institutional affiliations — and uses AI to (a) summarize that public information into a starting About statement (b) map each researcher's work to industry applications drawn from Halo's taxonomy with a confidence score and (c) analyze the aggregate scores across profiles to assess an institution's top application areas.

For faculty profiles, Halo is not creating biographical content from scratch; it's organizing public scholarly data so researchers are discoverable by industry partners. Pages are clearly labeled as unclaimed and AI-summarized until the faculty member has reviewed and confirmed them, and faculty can claim, edit, or request removal at any time. For institution pages, Halo is looking at a roll-up of faculty profiles.

The goal: give institutions and faculty an easier on-ramp to industry discovery, grounded in concrete published evidence that builds credibility with industry partners.

Faculty profiles

What data is used

Two sources, in this order of precedence:

1. Publication data from OpenAlex. OpenAlex is the open scholarly database that aggregates publication metadata (titles, abstracts, authors, journals, citations) from across the web. For each researcher, Halo pulls the publication records OpenAlex associates with that author. The AI reads titles and abstracts — not full papers.

2. Faculty-provided input. When faculty claim their profile and set their industry applications directly, their selections replace inferred classifications. Tags they keep from the AI's suggestions become faculty-confirmed.

What the AI does not use: full-text papers, grant data, patent filings, social media, news articles, or anything not in OpenAlex or directly entered by the faculty member.

Known limitations:

• Partial OpenAlex coverage. Some OpenAlex profiles contain only a subset of a researcher's publications — typically the most recent or most-cited. In those cases, the classifier reasons from the visible subset.

• Imperfect author disambiguation. The vast majority of researcher-publication matches are correct, but there's a non-zero risk that publications from a different person with the same name get merged into a profile, or that one researcher's work gets split across profiles. Halo flags incoherent publication sets as likely disambiguation issues and classifies conservatively in those cases. For errors that end users encounter, the fastest fix is the "report error" button on every page, accessible to faculty, university administrators, and industry users. Halo resolves disambiguation issues as they're reported.

What the AI does

For each researcher, the AI:

1. Reads the publication titles and abstracts as a single body of research.

2. Identifies the researcher's program of work — the coherent thread across their publications.

3. Creates a claimable profile and attaches the faculty member's recent publications.

4. Generates a starting About statement (2–3 sentences, industry-facing): broad and category-level rather than deep-technical, leading with career markers (patents, awards, institutions, years of experience) when present, or research domains and processes otherwise.

5. Maps the program of work to industry applications. The AI walks through Halo's taxonomy (~150 industry applications across 15 industry sectors) and identifies which applications the researcher's published work is positioned to support. The mapping is driven by the end-use of the research — the industry context where the work could be deployed — not by methodology, molecules, or vocabulary overlap. For example, a researcher developing better batteries for crop-monitoring drones is classified under Crop yield optimization (the agricultural end-use), not under Energy storage systems (the underlying technology).

6. Assigns each mapped application a confidence score from 1 to 5. Confidence 5 means the area is the researcher's primary research identity, with multiple supporting publications. Confidence 3 means a real but secondary research thread. Confidence 1–2 means thin or adjacent assignments. Only high-confidence assignments are displayed publicly; lower-confidence assignments are retained in Halo's data but not surfaced.

The AI is not doing: sentiment analysis, personality inference, prestige ranking, predicting future research direction, or drawing conclusions from anything beyond the publication trail and faculty input.

How accurate is the classification?

Halo evaluates classifier accuracy regularly against hand-scored samples with web-search and publication-level validation. A 300-user accuracy audit (June 2026) found that at the current display threshold, roughly 80% of classified profiles are accurately tagged (good/excellent), 10% are moderate, and 10% fail. The remaining failures are driven primarily by author disambiguation in upstream publication data — not by classification logic errors. The full audit methodology and results are available on request.

We deliberately prioritize precision over recall. Anyone looking at a profile or institution page should think "yeah, this is accurate — but you're missing this one thing" rather than "where did this come from?"

Unclaimed AI-summarized faculty profiles

Most faculty profiles on Halo start as unclaimed AI-summarized faculty profiles — generated from public OpenAlex publication data before the faculty member has interacted with Halo.

Unclaimed profiles are clearly labeled as such. Every unclaimed profile carries a visible "unclaimed" / "AI-summarized" indicator so anyone viewing the profile knows it has not been reviewed or confirmed by the researcher. Claimed profiles do not carry this label.

The unclaimed-by-default approach is intentional, for two reasons:

1. Institution pages need coverage to be meaningful. An institution's top-application signal only works if most of its faculty are represented. Waiting for every researcher to opt in would produce sparse, biased coverage that doesn't reflect actual research strengths.

2. Unclaimed profiles are a starting point, not a finished product. When a faculty member claims their profile, they arrive at something already populated — publications, About statement, starting industry applications — rather than a blank slate. They can accept, refine, add, or remove the AI's suggestions.

Faculty controls

• Claim and edit. Faculty can claim their profile and edit the headline, About section, and other fields.

• Override or add industry applications. Faculty can change their classifications directly after claiming. Faculty selections replace AI-inferred ones; changes propagate to institution pages within 5 minutes.

• Report errors. The "report error" button on every profile is the fastest path for disambiguation errors, publications attributed to the wrong person, or factual errors in AI-generated content. Halo resolves these as they're reported.

• Request removal. Faculty can request that their profile be removed from public display at any time.

Institution pages

How top industry applications are chosen

Each institution page surfaces the industry applications where the institution has the strongest research presence and the strongest evidence of industry collaboration. The blurb shown on the institution page:

Our AI pipeline uses publication data from OpenAlex to assess which industry applications have the highest concentration of publishing faculty at this institution, and surfaces those with the strongest evidence of industry collaboration.

How this works:

1. For each faculty member, an LLM assigns up to 8 industry applications that best characterize their work, with a confidence score from 0–5 per application (faculty self-identified applications receive the maximum score of 5). Only high-confidence classifications are used for institution-level ranking.

2. Halo aggregates these classifications across all of the institution's publishing faculty to produce a concentration score (headcount) for each industry application.

3. The top 20% of industry applications by concentration form the candidate pool.

4. Each candidate is scored for publicly discoverable evidence of industry partnership at the institution — named corporate partners, funded spinoffs, joint research agreements, and similar markers.

5. Candidates are re-ranked by industry partnership evidence, with concentration as a tiebreaker.

6. The top 8 are displayed on the institution page.

The concentration gate ensures every displayed application has real researcher mass behind it.

Classification coverage and precision. Around 75% of an institution's publishing faculty receive high-confidence classifications — these are the researchers whose industry applications are displayed on their profiles and who contribute to the institution's concentration ranking. The remaining ~25% are classified at lower confidence; their tags are retained in Halo's data but are not displayed and do not factor into institutional strengths. We deliberately prioritize precision over recall: fewer, more accurate tags rather than broader, noisier coverage. The partnership re-ranking ensures the institution's strongest industry stories lead the page. The partnership layer only affects ordering, not inclusion — no applications are knocked out based on partnership evidence alone.

Display vs. search. The confidence threshold governs what appears on researcher profiles and institution pages. Industry application tags are a small fraction of what Halo's network search uses to find good matches — the bulk of the matching signal comes from publications (titles and abstracts), solution listings, fundings, and patents. A researcher who doesn't display any industry applications on their profile can still surface in search results for a relevant query based on these other signals. The precision-over-recall tradeoff applies to display only; search coverage is not reduced.

University controls

• Self-serve via faculty. Faculty members can claim their profiles and update their own application areas, which will recalculate institutional top applications within 5 minutes.

• Customer success. Universities partnering with Halo can discuss which industry application terms appear on their institution page and which faculty are highlighted.

• Disambiguation and major issues. Halo resolves disambiguation errors on the back end as they're reported via the "report error" button on any page.

• Request removals on behalf of faculty. Universities can request removals on behalf of their faculty through their Halo account manager.

General policies

Removals and opt-outs

Halo treats removal requests as legitimate by default and processes them through three paths:

Self-serve corrections via claim. For most accuracy concerns — a wrong industry application, outdated headline, missing publication — the fastest fix is to claim the profile and edit it directly. Faculty edits take precedence over AI-inferred content, and changes propagate within 5 minutes.

Report-error button. For issues the claim flow can't fix — disambiguation errors, publications attributed to the wrong person, factual errors — every profile has a "report error" button accessible to faculty, university administrators, and industry users.

Formal removal requests. A faculty member can request that their profile be removed from public display by contacting support@halo.science.

University-level requests. Universities partnering with Halo can work with customer success on institution-level concerns. Universities not in a paid partnership can still request changes via the report-error mechanism.

Common questions

Q: What data is actually used to create my unclaimed profile?

Publication data from OpenAlex (titles, abstracts, author metadata) and, once claimed, faculty-provided input. See the "What data is used" section above.

Q: How accurate is the classification?

At the current display threshold, roughly 80% of classified profiles are accurately tagged (good/excellent), 10% are moderate, and 10% fail — almost all due to author disambiguation in upstream data. We deliberately prioritize precision over recall.

Q: Why was my colleague classified under X but I'm not, when we work on similar things?

Classification is per-researcher, driven by each researcher's own publication trail. Two researchers in the same lab can end up with different applications if their published outputs emphasize different end-use contexts. The fastest fix is to claim your profile and update your applications.

Q: Something on my profile isn't accurate. What do I do?

Claim your profile and update the field directly. Faculty selections replace AI-inferred ones. If there is an error you can't fix via claiming, click the "Report error" button so Halo can resolve it. You may also request complete removal.

Q: My publication record is incomplete on Halo — why?

Halo's publication data comes from OpenAlex, which has variable coverage. If you notice missing publications, you can add them directly after claiming your profile.

Q: Are unclaimed profiles publicly searchable before they're claimed?

Yes. They appear on halo.science, on their institution's page, and on industry application directory pages. The confidence display filter applies identically to both claimed and unclaimed profiles.

Q: Does Halo sell my data?

No. Halo's business is helping industry R&D teams find and partner with academic researchers — not selling researcher data. Profiles are publicly visible on halo.science so industry scouts can discover researchers for partnership conversations, but Halo does not license, resell, or syndicate researcher data to data brokers, recruiters, or other third parties.

Q: How can institutions onboard faculty quickly?

Paying institutions can onboard faculty using easy-claim links that Halo provides. The university admin can distribute these links directly or Halo can send them on the admin's behalf, after a heads-up email to faculty if preferred.