About this study
This AI Product Case Study walks through two AI tools I built and shipped: the Profile Optimization Tool and the Job Application Workflow Agent. Rather than just describing what they do, it shows the product thinking behind them: how I scoped the problem, picked the right user, mapped their journey, prioritized what to build, and defined what success looks like.
350+
Coaching sessions
2
Shipped AI tools
8
Product sense steps
Both tools grew directly from 350+ career coaching sessions, direct user observation, and my own active job search. Not hypothetical personas.
Step 1: Scope the problem
It started with a deliberately vague observation: job seekers are struggling with their applications. That's too broad to build anything meaningful, so the first step was to get specific.
| Question | Answer |
|---|---|
| What kind of struggling? | Not finding roles. Job boards handle discovery. The real pain is converting applications: getting past the first filter. |
| Where in the process? | Before applying (is my profile ready?) and while applying (how do I tailor for this JD?). |
| What channel? | Desktop web. Job seekers prep materials at a desk, not on their phone. |
| Core mechanism? | Resumes and LinkedIn profiles evaluated against a job description. The gap is the problem worth solving. |
Job seekers don't know how well their profile matches a specific role, and they don't have a fast way to close those gaps before they apply.
Out of scope
| Not solving | Why |
|---|---|
| Job discovery | LinkedIn and Indeed already solve this |
| Interview prep | Different tools, different moment in the journey |
| Salary negotiation | Happens after application |
| Auto-applying on behalf of users | Trust and accuracy risks are too high |
Step 2: Segment the users
| Segment | Who they are | Application behavior | Core challenge |
|---|---|---|---|
| Early-career (0-3 yrs) | Recent grads, first or second job | High volume, 20-50 apps/month | Generic resumes; don't know which keywords matter to recruiters or ATS |
| Mid-career (3-10 yrs) | Professionals pivoting roles or industries | Selective, 5-15 apps/month | Rich experience but struggle to translate it; bullets describe tasks not outcomes |
| Senior / career changers | Directors, VPs, or major domain switches | Very selective, 2-8 apps/month | Hard to condense scope; risk of appearing over- or under-qualified |
Segments based on behavioral patterns observed across 350+ career coaching sessions.
Step 3: Pick the target segment
Target: Mid-career professionals with 3-10 years of experience who are pivoting into a new role or industry, e.g. an operations manager moving into product management.
| Reason | Detail |
|---|---|
| Highest pain-to-value ratio | Experience exists. It needs reframing. AI is strong at translation without inventing facts. |
| Most common in coaching | This group came up most often for tailoring help and bullet rewrites. |
| Invests time per application | Selective applicants who want real return on tailoring effort. |
| Outcomes are measurable | Fit scores before/after, completion rates, acceptance rates. |
Step 4: Persona & journey
Meet Kathy, the frustrated career pivoter
| Attribute | Detail |
|---|---|
| Background | Five years in operations at a SaaS company. Vendor workflows, cross-functional projects, JIRA and Notion |
| Goal | Move into a Product Manager role at a tech company |
| Behavior | 8-12 PM roles/month; 30-45 min tailoring each application; rarely hears back |
| In her own words | "I know I have the skills. I've done PM work. But I have no idea why I'm not getting calls." |
Kathy's seven-stage journey
| Stage | Pain point | Severity |
|---|---|---|
| Discover role | No quick way to validate fit before committing time | High |
| Assess fit | Slow, subjective comparison; unsure which keywords matter to ATS | High |
| Tailor resume | Bullets describe tasks not outcomes; gives up and sends same resume | High |
| Update LinkedIn | Profile static and generic, not positioned for target roles | Moderate |
| Write cover letter | Generic opening; no connection to specific JD | Moderate |
| Submit | No confidence she put her best version forward | Low |
| Wait | No feedback loop after rejection | Low (out of scope) |
Step 5: Prioritize pain points
| Pain point | Priority | AI suitability |
|---|---|---|
| Can't quickly tell if she's a fit before tailoring | HIGH | High. Keyword pattern-matching is core LLM strength |
| Can't rewrite bullets in JD language | HIGH | High. Grounded in user's existing text, low hallucination risk |
| LinkedIn not positioned for target role | MODERATE | High. Once per campaign, not every application |
| Cover letter from scratch every time | MODERATE | High. Best after fit analysis and rewrites |
| No feedback loop after rejection | LOW | Low. Outside our control |
Top pains to solve: (1) Instant fit assessment with grounded JD citations, (2) Bullet reframing in outcome language, (3) LinkedIn audit as a complementary baseline check.
Step 6: List the solutions
| Feature | Pain it solves | How it works |
|---|---|---|
| Fit score engine | Can't tell if she's a fit | 0-100 score with tier, keyword breakdown, and JD line citations |
| Skill gap analysis | Doesn't know what's missing | Required skills from JD vs. resume with evidence attached |
| Bullet rewriter | Can't reframe in JD language | Rewrites each bullet; user accepts or rejects one by one |
| Cover letter generator | Starts from scratch | Streams a letter built from fit analysis and accepted rewrites |
| LinkedIn section audit | Profile not positioned | Scans headline, About, Experience, Skills with JD-tied fixes |
| Match report & print | No record of submission | Exportable report: score, keywords, actions, print to PDF |
Step 7: Prioritize features
| Feature | Ship priority |
|---|---|
| Fit score & gap analysis | SHIP FIRST. Entry point; needs only resume + JD |
| Bullet rewriter | SHIP SECOND. Conversion step; grounded in user's text |
| Cover letter generator | v1. Downstream of fit + rewrites (Step 4) |
| LinkedIn section audit | v1 parallel. Separate moment, Profile Optimization Tool |
| Match report & print | v1 bundled. Needs score first |
The cover letter ships in the Workflow Agent because it's downstream of fit analysis and rewrites. The LinkedIn audit lives in Profile Optimization because it serves a different trigger: baseline readiness, not active application.
Step 8: Wireframes & success metrics
Profile Optimization Tool
Solves LinkedIn/resume baseline readiness. Score-first layout: match %, then gaps, then actions. Nothing stored between sessions. A deliberate trust decision.
| Metric | Type | Target |
|---|---|---|
| Analysis completion rate | Primary | > 85% |
| Time to report (p50) | Primary | < 60 seconds |
| Action plan engagement | Secondary | > 60% scroll past score |
| Print / export rate | Secondary | > 20% of sessions |
| AI hallucination rate | Guardrail | 0 per 100 analyses |
Job Application Workflow Agent
Four-step guided workflow: Analyze Fit, Surface Gaps, Rewrite Bullets, Draft Cover Letter. Accept/reject on every bullet. Streaming cover letter. Golden-set regression before every ship.
| Metric | Type | Target |
|---|---|---|
| Workflow completion (all 4 steps) | Primary | > 60% |
| Fit score accuracy (golden set) | Primary | 5/5 per deploy |
| Bullet acceptance rate | Secondary | > 50% of rewrites accepted |
| Cover letter generation rate | Secondary | > 40% who reach Step 3 |
| Step 1 to Step 2 drop-off | Secondary | < 20% |
| Hallucination rate | Guardrail | 0 per 100 analyses |
Summary: how I think about product
| Framework step | Decision made |
|---|---|
| Scope the problem | Narrowed to closing the gap between profile and a specific JD |
| Segment users | Three segments by experience level and application behavior |
| Pick target segment | Mid-career pivoters. Highest pain, best AI fit, largest in coaching data |
| Persona & journey | Kathy's seven-stage journey maps real pain at every touchpoint |
| Prioritize pains | Fit assessment and bullet rewriting ranked highest |
| List solutions | Six features, each tied to a specific pain. Nothing without justification |
| Prioritize features | Fit score first, bullet rewriter second, LinkedIn audit in parallel |
| Wireframes & metrics | Primary metrics per tool + hallucination guardrails |
The persona, pain points, and feature priorities are not hypothetical. They come from 350+ career coaching sessions, direct observation, and my own active job search.