What the AI is actually doing

Before anyone on your team opens a candidate profile, Talinty's AI has read the resume, scored it against the role, compared it to the criteria you set, and written a structured analysis. The AI's job is to do the first pass; yours is to do the second pass on a much shorter list.

Inside a candidate profile, the AI's output shows up in three places: the scores at the top of the Summary, the AI Vetting Brief below it, and the AI hiring recommendation that surfaces when the candidate is ready for a decision.

The two scores

Every candidate gets two numbers, side by side:

• Resume match score. How closely the resume aligns with the job description. Weighted toward the criteria you set as required. A 92 means strong overall alignment; a 68 means partial match with notable gaps.

• AI skill match %. A second pass that looks at actual skill alignment beyond keyword matches. Where the resume score measures fit on the surface, the skill score measures fit underneath. It draws on the parsed resume data plus any assessments the candidate has completed.

Two scores instead of one is deliberate. A candidate can score 90 on resume match (good keywords, clean structure) and 55 on skill match (the actual skills tell a thinner story). Both numbers matter. Either alone would mislead.

The three-column vetting brief

Click into the AI Vetting Brief section and you'll see three columns:

• Why they match. The case for this candidate, written in plain language. Specific phrases from the resume, named skills, concrete experience that maps to the role.

• Bonus strengths. Things the candidate brings that weren't on the original criteria list but might matter. Relevant adjacent experience, secondary skills, language fluency, certifications you didn't ask for but might value.

• Things to consider. The gaps, risks, and questions worth following up on. Years of experience light, missing specific technologies, employment gap with no context, geographic constraint. Not deal-breakers; just the parts you'd want to ask about.

The brief is generated per candidate per role. The same candidate in two different roles gets two different briefs, because the criteria are different.

The AI hiring recommendation

When a candidate has gathered enough signal (resume, assessment scores, video interview, team scorecards), the AI hiring recommendation appears in the actions panel: Advance, Hold, or Future Roles.

The recommendation isn't a vote count of the scores. It's a weighted synthesis of everything Talinty has seen: AI scores, assessment results, video interview transcripts, your team's notes and scorecards. Strong scorecards from people who know the role pull more weight than a casual note from someone tangentially involved.

Clicking the recommendation acts on it (advancing the candidate, placing a hold, or moving them to the talent pool). You can override at any time.

When you think the AI is wrong

The AI sometimes scores a candidate higher or lower than your gut. A few patterns:

• The score is high but the brief reads thin. Often a sign that the resume is well-formatted (lots of right keywords, clean structure) but lacks substance. Read the Things to consider column carefully; it usually catches this.

• The score is low but the candidate looks strong. Often a parse issue (the resume is heavily designed, image-based, or in a layout the parser struggles with) or a criteria mismatch (the job description is too narrow). Check the parsed-data view of the resume to see what the AI actually extracted. If it's wrong, edit the fields and the score updates.

• The score is fine but you disagree on fit. This is what scorecards are for. Your judgment adds to the AI's evaluation; it doesn't replace it. Write the scorecard and explain your reasoning. The recommendation will reflect what you said next time it refreshes.

You're the one who decides. The AI gives you a defensible starting point and the reasoning behind it. The decision is still yours.

What the AI deliberately doesn't do

A few things worth knowing:

• The AI doesn't auto-reject candidates. Disqualification is always a human action with a recorded reason.

• The AI doesn't read protected characteristics (age, gender, ethnicity, marital status) into the score. The model is instructed to ignore them, and the workspace audit trail records every decision for review.

• The AI doesn't crawl the internet for additional information about a candidate. It works only with what's in the profile.

• The AI doesn't make irreversible moves. Every action it takes (or recommends) is reviewable, overrideable, and logged.

Configuring what the AI weights, when it triggers, and where its recommendation thresholds sit lives in the Working with the AI and Tuning the AI categories.