v0.17.0 LLM live: OpenAI · gpt-5-mini
How this works. Official sources, checked answers.

Under the hood

Where these answers come from

Career Explorer doesn't guess. Every recommendation it makes is built on the federal government's own published rules: OPM's occupational series definitions, OPM's qualification standards, the Code of Federal Regulations, and live USAJOBS listings. AI helps read your resume and translate government language into plain English - but it works from these official sources, and the answers it gives are checked against them before you see them.

Three separate questions, never blended

Federal hiring really asks three different questions, and this site keeps them separate on purpose:

  1. Can you apply at all? (Eligibility) - determined by hiring paths you tell us about: veteran, current federal employee, recent graduate, and so on. You answer this; we never infer it.
  2. Where does your background fit? (Series fit) - your experience matched against all 618 the government uses to classify white-collar work.
  3. Would you meet the minimum qualifications? (Qualification) - your resume checked against a specific announcement's published requirements, the way a federal HR specialist would read them.

Many job tools collapse these into one score. We don't, because federal HR doesn't: you can be a perfect professional fit for a job you're not eligible to apply for, and eligible for one you don't yet qualify for. Knowing which problem you have is most of the battle.

The three questions kept separate: 1, eligibility - can you apply at all, decided by hiring paths you tell us about. 2, series fit - where your background fits among the government's occupational series. 3, qualification - whether you meet a specific announcement's published minimum requirements.

The official sources we build on

For veterans specifically: we hold 1,359 military specialties across all 6 service branches - every one with a plain-language description of the work - connected to civilian job titles by 13,394 individual links. That's roughly ten civilian job titles for every specialty, which is the point: your service didn't map to one civilian job, it mapped to a dozen you may never have heard called that.

We keep dated copies of the reference material (the qualification standards on this site were fetched 2026-08-13) and check the sources for changes on a schedule - so we can tell you exactly which version of a rule an answer was based on, and notice when OPM or the CFR changes something. The plain-language glossary covers 112 terms, each linked to the authority behind it.

Go read them yourself: Handbook of Occupational Groups and Families OPM General Schedule Qualification Standards OPM Title 5, Part 300 — employment (incl. time-in-grade) eCFR Title 5, Part 511 — position classification eCFR USAJOBS USAJOBS Military Crosswalk Search O*NET / U.S. DOL mos.directory mos.directory

The official sources stack: OPM's handbook covering every occupational series, OPM's General Schedule qualification standards including mandatory education rules, seven parts of Title 5 of the Code of Federal Regulations, veteran crosswalk records from mos.directory and O*NET, and the site's plain-language glossary.

What the AI does - and what it isn't allowed to do

AI (currently OpenAI models) is used for reading, not deciding:

Guardrails, enforced in code rather than requested politely:

What the AI does versus what it is prevented from doing: it reads resumes, translates announcements, and applies a written qualification procedure; it cannot invent a series match without supporting evidence, cannot supply a legal citation or URL, cannot decide your eligibility, and cannot return an unhedged 'qualified' verdict.

How each of the three checks actually runs

Each step below is tagged with who does it: you, deterministic code, or the AI. The AI is one tagged step in the middle of each flow, with code on both sides of it.

Eligibility flow: you state your hiring paths, code compares them against the announcement's paths, and a verdict is returned. No AI step appears anywhere in this flow.
Gate 1 - eligibility. Your answers, compared in code. No AI involved at any point.
Series fit flow: you upload a resume, code extracts the text, the AI proposes candidate series and structured skills, code then verifies each proposal against the official OPM series text, demotes implausible series-and-grade pairings, and ranks what survives - each match shown with the evidence behind it.
Gate 2 - series fit. The AI proposes; official OPM text verifies. A match the evidence can't support never reaches you.
Qualification flow: code assembles the announcement text, your resume profile, the vendored OPM standard and the time-in-grade calculation; the AI applies the written HR procedure; code then verifies every citation against the official text, drops unverifiable ones, and constrains the verdict to a hedged set.
Gate 3 - minimum qualifications. Official inputs assembled in code, the procedure applied by AI, every citation verified server-side before you see it.

What happens to your resume

Your resume is analyzed in memory and never written to disk. The extracted text is sent to OpenAI to build your profile (OpenAI is instructed not to store it), the original file is discarded as soon as the text is read, and the derived analysis is erased when your session ends. Visits are counted on our own server - no third-party script, no IP addresses, no way to identify you. The full details are on the Privacy & cookies page.

Your resume's journey: uploaded, held in memory, text extracted and the original file discarded, extracted text sent to the AI with a no-storage instruction, the derived analysis kept only for your session, then erased. Nothing is written to disk at any stage.

The honest limits

Want the version with file paths, counts, and provenance for every claim above? It's docs/ARCHITECTURE.md in the project's source repository. Military skill data courtesy of mos.directory; military crosswalk data from O*NET, U.S. Department of Labor, used under CC BY 4.0.