Career Pathway Matching

A career pathway match is defensible only when a counselor can see why this graduate was paired with this posting. The output names the skills profile and the job posting, then lists overlapping skills with source text from both sides. If either side is thin, the result stays empty. Empty is a quality outcome. Inventing a job, completing missing skills from a model prior, or stretching a public opening into a partner role is not.

This page is for career-services and placement counselors who already sit between a graduate's record and an employer inventory. Matching is retrieval plus explanation. It is not a placement. The counselor still owns the recommendation the student hears.

When the profile is a competency transcript rather than a résumé dump, cites are easier to audit because each claimed skill points at coursework, assessments, or documented experience. When the posting is a partner role already on file, the counselor can verify that an employer actually asked for those skills.

Do not start from an open web scrape. Start from two bounded sets.

The graduate side is the skills profile: competencies, credentials, work samples, and counselor notes already in the student record. Typical sources are the student information system (Ellucian-class), the learning management system (Canvas-class course and assessment history), and the career-services CRM that already holds employer notes (Salesforce-class). The profile is only as strong as those systems. If the record is a degree title and three internships with no skill language, matching has almost nothing to cite. Enrich the profile, or leave the match empty.

The employer side is the posting inventory the institution is allowed to use: partner jobs, internships, and graduate roles the career office has on file, including postings stored in Handshake-class career platforms. Handshake here means the class of campus career systems that already store employer partners and student-facing jobs, not a named product capability. If a role is not in that inventory, it is out of scope. High similarity to a posting the office does not represent is a failure, not a win.

Load both sets in the same session: student identifier, profile version, and a dated snapshot of eligible postings. Freeze the posting set for that run so a counselor is not reviewing a match against a job that vanished between ranking and advising.

Embedding-based matching compares the profile text to each posting in the frozen inventory. The score is a retrieval signal. The artifact the counselor needs is the cite pack: which profile passages support the overlap, which posting sentences asked for those skills, and which required skills have no supporting passage.

A ranked list without those cites is a leaderboard, not a match. Prefer evidence coverage over raw similarity. A role that shares named competencies with quoted posting language outranks a role that merely sits nearby in embedding space.

Illustrative example: a graduate in applied statistics has a profile that cites a capstone using R, a practicum cleaning survey data, and a methods course covering experimental design. The inventory includes a partner research-assistant posting that asks for R, survey data, and documentation of analysis choices, and a partner marketing-analyst posting that asks for paid-media bidding and dashboard ownership. The first role can be shown with quotes from the capstone and the practicum next to the posting's R and survey lines. The second role should not be dressed up as a pathway, because the profile does not contain paid media. If the office has no research-assistant posting on file, neither role is a match, even if similar jobs exist on the public web.

Show gap analysis on the same screen as the match: required skills with cites, preferred skills with cites, and required skills with no cite. Gaps are absences in the profile relative to that posting. They are not a generated training plan. If the posting never mentioned stakeholder management, do not add it as a gap. If the profile never mentioned Python, do not invent a Python credit to close a gap.

Thin inputs are common at graduation: late transcripts, internships described in one line, postings that are a job title and a salary band. The correct behavior is no match, with a reason the counselor can act on: profile too sparse, posting too sparse, or no eligible posting in inventory. That reason should point at the missing record, not at a suggested job.

Do not backfill from job market signal monitoring into the match itself. Labor-market signals can inform which partner conversations the office starts next term. They must not mint a fictional opening so a graduate has somewhere to go. Pathway matching answers a narrower question: given this profile and this inventory, which on-file roles have cited overlap, and where does the profile fall short of those postings.

If the only candidates are weak similarities to jobs the employer list does not contain, return empty and record that the inventory, not the student, is the constraint. That record is useful for employer-relations work. It is not a student-facing pathway.

Three errors show up as a polished list.

Matching a job the employer list does not contain. Similarity will retrieve public postings or stale feeds. If the office cannot name the partner contact and the posting record, suppress the result. Presenting an off-inventory job as a pathway creates a placement expectation the institution cannot honor.

Inventing a skill. Models complete patterns. A statistics graduate plus a data-adjacent posting often yields Python or SQL in the overlap list even when neither document contains those words. Every overlap line must quote the profile and the posting. If a quote cannot be produced, the skill is not in the match.

Treating a match as a placement. A ranked role is a conversation starter for advising, not an offer, not a referral, and not an employment outcome. Employment outcome tracking records what actually happened after counseling. Mixing the two inflates placement counts and trains staff to chase scores instead of employer follow-up.

A quieter error is showing the student the ranked list before a counselor has reviewed cites and gaps. Students read rank as destiny. Keep the first pass in the counselor workflow, such as a placement counselor co-pilot session, until a human has accepted, edited, or discarded the match.

The counselor checks three things before anything reaches the student: the posting is still in inventory and current, the cites are real, and the gaps are honest. They may demote a high-scoring role because of location, work authorization, hours, or a known employer issue that never appears in embeddings. They may promote a lower-scoring partner role because the office already has a hiring manager relationship. Those overrides belong in the advising note.

If review finds a fabricated skill or an off-list job, the match is rejected, not edited into a prettier story. If review finds a thin profile, the next step is profile work: transcript detail and project write-ups, not a weaker posting. If review finds a thin posting, the next step is employer-relations clarification of the role, not a guessed skill list.

Only after that review does the counselor decide what to recommend, what to frame as a stretch with named gaps, and what to withhold. The system produced evidence. The counselor produced the pathway conversation.