Layered Compass role intake and candidate evaluation interfaces

Case study

AI Hiring Copilot

An AI-native hiring system from role intent to candidate proof

at SeekOut ↗

Choose your depth

The problem, my contribution, outcomes, and strongest screens.

My role

Lead product design · AI experience strategy · Multi-sided service design

Product focus

An AI-native hiring system connecting role intent, ATS-synced evaluation, and reusable candidate proof across SeekOut's Connect product.

The Short Version

Hiring is usually three separate disasters wearing a trenchcoat: a hiring manager guessing at what they actually want, a recruiter drowning in a Greenhouse queue, and a candidate re-typing their resume into yet another form. Compass ties those three jobs into one system: a hiring manager's vague brief becomes a real rubric, a recruiter's queue becomes something Sam (the AI evaluator) can actually chew through, and a candidate walks away with a profile that outlives the one application they used it for.

The Mess

Every handoff in hiring is where context goes to die. A hiring manager's priorities get flattened into a job description, the job description gets skimmed by a recruiter triaging hundreds of these, and a candidate gets judged by a process nobody explains to them. Layering AI on top of that doesn't fix it. It just makes the black box faster. The real challenge was building something people would trust enough to hand real decisions to: criteria a hiring manager didn't need a machine-learning degree to set, automation a recruiter could watch working instead of taking on faith, and a screening experience a candidate could read as coaching instead of judgment from a robot they'll never meet.

My Part in This

  1. Designed the service end to end: hiring manager, recruiter, and candidate aren't three products wearing a shared logo, they're one system.
  2. Turned a pasted job description into structured role data, an editable rubric, and screening questions Sam drafts and a human actually approves.
  3. Mapped ATS stages explicitly, so Connect's automated actions sync back to Greenhouse predictably instead of recruiters finding out the hard way.

What Got Better

  1. Replaced four disconnected steps (intake, rubric, scoring, screening) with one role model that actually holds together.
  2. Gave recruiters a live, filterable path from hundreds of Greenhouse candidates to recommendations they can actually inspect and verify.
  3. Made ATS mapping and rubric approval explicit, deliberate steps: automation only does what a recruiter actually signed off on.
  4. Built lasting candidate value out of resume coaching, a reusable AI-forward profile, and a setup checklist that doesn't vanish after one application.

Impact

What Actually Moved

For the Business

Turned four separate operating headaches (role definition, ATS evaluation, collaborative review, candidate proof) into one governed model, instead of four teams quietly building four workarounds.

For the Humans

Hiring teams move faster without handing judgment to a black box, and candidates leave with clearer expectations and a profile they actually control, not just a 'thanks, we'll be in touch.'

For the Spreadsheet

Extends product value across hiring managers, recruiters, reviewers, and candidates: more surfaces for adoption and enterprise expansion, without inventing a metric nobody can verify.

Three connected journeys

Three Jobs, One System

Follow Compass from a hiring manager's first guess at what they want, through a recruiter actually running the evaluation, to the candidate on the other end of it. Same rubric, same AI evaluator, three very different jobs to get right.

Hiring manager flow

Turn Vague Into Verifiable

Hiring managers move from a pasted job description to a reviewed rubric and confirmed screening questions, choosing conversation or direct editing at every step, and never getting a surprise from something Sam did without asking first.

  1. Turn a pasted job description into structured role data Sam can evaluate against.
  2. Use a live conversation with Sam to surface priorities and dealbreakers a form would miss.
  3. Review, edit, and add to AI-drafted screening questions before they reach a candidate.
Hiring manager · 01
Choosing hiring manager at sign-in sets the entire onboarding path: no generic AI settings screen making you guess what you're even configuring.
Hiring manager · 02
Account creation sits right beside the product's own pitch, in case you forgot mid-signup why AI-powered screening was supposed to help you.
Hiring manager · 03
Job setup starts with the description itself. Sam starts reading it the moment it's pasted in, progress bar and all.
Hiring manager · 04
Evaluation hands back structured role data (title, location, skills, requirements, responsibilities), ready to review before anything moves forward: good, because this particular pass read a product design brief and confidently drafted a software engineer's skill list.
Hiring manager · 05
Sam drafts three screening questions straight from the job description, each labelled AI generated and fully editable, because 'trust me' isn't a UI pattern.
Hiring manager · 06
A hiring manager can add a fourth question directly, or describe what they actually want and let Sam take a swing at drafting it.
Hiring manager · 07
When intake moves to conversation, Sam asks about priorities and dealbreakers, the stuff that usually only comes up after the wrong candidate is three interviews deep.
Hiring manager · 08
The conversation resolves into concrete screening questions the hiring manager actually confirms, before any of them reach a candidate.

Hiring manager flow

Hiring manager flow

Turn Vague Into Verifiable

Hiring managers move from a pasted job description to a reviewed rubric and confirmed screening questions, choosing conversation or direct editing at every step, and never getting a surprise from something Sam did without asking first.

  1. Turn a pasted job description into structured role data Sam can evaluate against.
  2. Use a live conversation with Sam to surface priorities and dealbreakers a form would miss.
  3. Review, edit, and add to AI-drafted screening questions before they reach a candidate.
Hiring manager · 01
Choosing hiring manager at sign-in sets the entire onboarding path: no generic AI settings screen making you guess what you're even configuring.
Hiring manager · 02
Account creation sits right beside the product's own pitch, in case you forgot mid-signup why AI-powered screening was supposed to help you.
Hiring manager · 03
Job setup starts with the description itself. Sam starts reading it the moment it's pasted in, progress bar and all.
Hiring manager · 04
Evaluation hands back structured role data (title, location, skills, requirements, responsibilities), ready to review before anything moves forward: good, because this particular pass read a product design brief and confidently drafted a software engineer's skill list.
Hiring manager · 05
Sam drafts three screening questions straight from the job description, each labelled AI generated and fully editable, because 'trust me' isn't a UI pattern.
Hiring manager · 06
A hiring manager can add a fourth question directly, or describe what they actually want and let Sam take a swing at drafting it.
Hiring manager · 07
When intake moves to conversation, Sam asks about priorities and dealbreakers, the stuff that usually only comes up after the wrong candidate is three interviews deep.
Hiring manager · 08
The conversation resolves into concrete screening questions the hiring manager actually confirms, before any of them reach a candidate.

Recruiter flow

Recruiter flow

Move Fast, Keep Receipts

Recruiters connect a live Greenhouse job, map its stages to Connect, approve the rubric Sam builds from intake, then watch and filter AI scoring across hundreds of candidates in real time, with every automated action traceable back to a stage the recruiter actually defined.

  1. Map Greenhouse stages to Connect so automated actions sync predictably back to the ATS.
  2. Approve the evaluation rubric as a deliberate step before Sam scores a single candidate.
  3. Filter hundreds of scored candidates down to the ones worth a closer look.
Recruiter · 01
Recruiters land on the same entry screen but pick their own lane (hiring manager or recruiter) out of one shared product.
Recruiter · 02
Before setup even starts, the product states its trade plainly: a two-minute conversation, in exchange for AI taking the first pass on every applicant.
Recruiter · 03
An empty Jobs dashboard still earns its keep, explaining what happens next and how long setup takes instead of just sitting there blank.
Recruiter · 04
Recruiters choose how they're sourcing for a role (network referrals or inbound ATS candidates) and can run both later, no commitment required upfront.
Recruiter · 05
Connecting a job pulls the live list straight from Greenhouse, flagging which postings are new, already set up, or already mid-flight.
Recruiter · 06
Selecting a posting keeps the full Greenhouse context (status, candidate count, owner) visible through the rest of setup, so nothing gets lost in translation.
Recruiter · 07
ATS stage mapping makes the sync explicit: which Greenhouse stage starts Sam, and exactly where advance, AI-screen, and reject actions move a candidate.
Recruiter · 08
Intake offers a voice or text conversation with Sam, or a direct skip straight to a rubric drafted from the job description alone, for recruiters who'd rather not small-talk with an AI.
Recruiter · 09
Before a voice intake starts, the product sets expectations on time, environment, and what happens if the call drops, because nobody likes finding out the hard way.
Recruiter · 10
The rubric Sam built from intake and the job description gets reviewed, edited, and reordered. Approving it is a real decision, made once and on purpose.
Recruiter · 11
A live job view shows Sam scoring candidates against the rubric in real time, funnel counts and status updating as the work happens.
Recruiter · 12
A job-details panel keeps the approved rubric, ATS configuration, team members, and interview assets one click away, so nobody has to go hunting mid-review.
Recruiter · 13
Filters narrow by must-haves, nice-to-haves, reviewer, decision, and stage, precise enough to isolate exactly the candidates worth a second look.
Recruiter · 14
One filter takes 972 candidates down to 79, without losing the scoring, summary, or status columns that made the list useful in the first place.

Candidate flow

Candidate flow

Make Evaluation Less Mysterious

Candidates upload a resume and get prioritised, specific feedback before building an AI-forward profile (skills, career history, an introduction video) that outlives the one application they made it for. A Chrome extension then carries that profile into the application itself, autofilling routine fields while keeping tailored content and unanswered questions visible.

  1. Turn resume feedback into prioritised critical, urgent, and optional fixes.
  2. Guide candidates through recording a short introduction video for their profile.
  3. Leave candidates with an editable, shareable profile instead of a one-time application artifact.
  4. Use the Chrome extension to autofill applications without hiding tailored documents, missing answers, or candidate-controlled fields.
Candidate · 01
Profile creation starts with a resume upload alongside GitHub and portfolio links, the raw material Sam actually has to work with.
Candidate · 02
When a resume is already strong, Sam just says so. It doesn't manufacture feedback to justify the step existing.
Candidate · 03
Weaker resumes get specific, counted feedback (critical, urgent, optional) instead of one mystery score you're left to interpret alone.
Candidate · 04
Resume analysis stays visible while Sam works, then hands straight off into building a tailored profile template.
Candidate · 05
Candidates get a script, on-camera guidance, and a countdown before recording an introduction video, because nobody likes staring at a blank record button in silence.
Candidate · 06
The finished profile presents skills and career journey together, with a setup checklist tracking what's left, the good kind of nagging. Every field stays editable for a reason: this pass wrote the headline as a product designer and the summary as a software engineer, and somebody has to pick a lane.
Candidate · 07
Publishing explains what sharing actually means (a unique, trackable link per job application) before the candidate commits to hitting send.
Candidate · 08
Candidates can add, edit, or reorder roles directly inside the profile, without ever leaving the page.
Candidate · 09
Skills are a simple, removable tag list the candidate controls, sitting right next to the AI-assembled version they can just fix.

Chrome extension

Take the Profile to the Application

The candidate journey continues on the job application itself. Connect autofills repeat information from the candidate's profile, tailors the resume and cover letter, and keeps errors visible for the candidate to resolve instead of quietly guessing.

Candidate · 10
The side panel narrates autofill while it works, then returns a compact checklist of what was tailored, what was completed, and which questions still need the candidate.
Candidate · 11
Expanded states expose the tailored cover letter and every profile-supplied field, so convenience never turns into invisible automation and candidates can inspect exactly what will be submitted.

Continue reading

There Is More. Naturally.