All projects
PERSONAL BUILD · AI PRODUCT · 2026

The AI PM Brain.

I am the user. I built the system I needed to become a recognized AI PM, and I use it every day.

49
posts audited
11
voice checks per draft
6
job boards scanned daily
17
week sprint
01 · Problem

The problem

I was running a 17-week plan across LinkedIn content, study, and a job search. Notes, drafts and applications lived in six places. I wanted one system that told me what to do today and showed me whether it worked.

02 · Product

What it does

  • Today: the one list of what to do, from posts due to follow-ups.
  • Content: a weekly batch of seven posts planned and drafted from my own work, checked before I see them.
  • Study: concept notes and a 17-week plan.
  • Job: a daily scan of open product roles, a fit score for each, and a tailored CV and cover letter.
  • Grow: research on creators and posts, with a score for how far a post beat its author's normal.
  • Track: weekly review of actuals against targets.
  • Settings: sprint dates and AI cost limits.
03 · Decisions

Decisions I made

Build the part nobody was building.

Existing tools did research or scheduling. None checked a draft against my own voice. I dropped a copy of a research tool and built the drafting and checking layer instead.

Treat the voice rules as an eval.

Every draft is scored against 11 fixed rules. The same rules rank outputs from different models, so "which model" becomes a measurement, not an opinion.

Let my own data change the plan.

I audited my last 49 posts. The median got 615 impressions. Explainers, half my output, sat near 547. A post about a build that broke got 5,040. I moved the plan to three proof-of-work posts a week. [Add the result after four weeks of data.]

Stop the tool from making things up.

Drafts and CVs flag any number or employer that is not in my resume or stories. Missing facts stay as [fill] instead of invented text. The app never submits an application. I send it myself.

Cap the cost and park the money.

Scraping has a monthly hard limit. Every login fails closed. Selling it is parked until a user can bring their own keys and own their data.

04 · Build

How it is built

Next.js and React. No database: the app saves to a private GitHub repo, so every change is a commit. Claude writes drafts and reviews. Scheduled jobs run the weekly batch and the daily role scan on Vercel.

05 · Results

Results

[Add 2 to 3 real numbers once the sprint has data: posts shipped, median impressions before and after, applications sent.]

06 · Screenshots

Screenshots

Screenshot placeholder
Screenshot placeholder
Screenshot placeholder
Screenshot placeholder

Screens use sample data.

07 · Want one?

Want one?

The app is built for one person today. If you want a starter version of the system, join the list and I will send it when it is ready.