I just launched JobPocket — a browser extension that scores your resume fit and tailors it with AI, amongst other things

Hey IH! Just shipped my first product — a browser extension that tailors your resume to job postings with AI. Wanted to share the journey and get feedback.

The problem I was solving for myself

I was doing a spray-and-pray job search while also trying to network. The networking is what actually lands roles, but every time I wanted to tailor a resume I'd lose 10-15 minutes alt-tabbing between the job post, ChatGPT, my resume file, a Word doc — copy-pasting back and forth, reformatting, exporting. Death by a thousand tabs.

I tried existing tools but they were all these heavyweight platforms with onboarding flows and dashboards and profile builders. I didn't want a system. I wanted a button on the job page that just does the thing so I can get back to networking.

What I built

JobPocket is a browser extension (Chrome + Firefox) that sits on the job listing page itself. No tab switching, no copy-pasting, no separate app.

The workflow is:

Open a job on LinkedIn, Indeed, or Glassdoor
Click "Extract Job" — it pulls the title, company, and full description automatically
Click "Score Fit" — get a 0-100 match score, a hire/pass recommendation, and a breakdown of your strengths vs gaps for that specific role
Click "Tailor Resume" — AI rewrites your full resume aligned to the job's language and priorities
Edit if you want (built-in markdown editor), then download as DOCX
The extension also tracks every job you've scored and lets you mark when you've applied. Your whole pipeline lives in the sidebar, stored locally in your browser.

The scoring step is actually the part I use most. It saves me from spending time tailoring for a role that's a 35/100 fit anyway. Triage first, then invest time only in the good matches. The whole loop takes about 30 seconds per job.

The local AI angle

This is what I'm most excited about technically. You can run the whole thing through Ollama on your own machine. Your resume and job data never hit any external server.

Default models:

Scoring: qwen3.5:0.8b (tiny, fast — runs on basically anything)
Tailoring: qwen3:4b (still small enough for most laptops, good structured output)

You can swap these for any Ollama-compatible model in settings. I spent a lot of time getting small models to produce clean, structured output — no reasoning preambles, no markdown wrapped in code fences, consistent JSON. It's surprisingly hard to get tiny models to just follow a format. Ended up using structured output schemas, /no_think flags, and post-processing to catch when models go off-script.

There's also a hosted mode (GPT-4o / GPT-4o-mini) for people who don't want to deal with local setup. Pay-per-use, no subscription.

Tech decisions

No framework — vanilla JS, no React, no build step. Extensions have weird constraints (CSP, service workers, cross-origin restrictions) and I
didn't want a bundler adding complexity on top
Supabase for auth and credit tracking
Stripe for payments
Deno edge functions for the hosted AI path
Ollama for local inference

Total infra cost right now: ~$0/month (Supabase free tier, edge functions only run when called)

Revenue model

Free to install, 5 credits included
Credit packs: $5 (25 credits), $10 (60), $25 (175)
Everything is 1 credit — score, tailor, whatever. Simple.
Local mode: $25 one-time unlock — unlimited forever, no credits needed

I deliberately avoided subscriptions. Job searching is temporary. Nobody wants another $9/month hanging around after they land a role. Pay for what you use, or pay once and own it.

What I'm unsure about

Is $10 for lifetime local mode underpriced? It's my best feature but also a one-time sale with no recurring revenue
Should I offer a free tier with limited tailoring (like 2 free tailors/month) to get more people in the door?
How do you market something like this without being spammy in job-seeker communities?