Network Growth Assistant

How it works

1. 01

   You add people by pasting a search page or importing a file.

2. 02

   Each day the app sorts the new people and builds your list for the day.

3. 03

   You open each profile and send the invitation yourself, then mark it as sent.

4. 04

   When someone accepts, you record it, and the app creates the follow-up reminders for that person.

5. 05

   The overview shows how close you are to your goals. Only accepted invitations count.

6. 06

   You search for jobs, save the ones you like, track your applications, and prepare for interviews.

Screenshots

Why

Networking tools tend to flatter you with vanity numbers or automate the part that should stay personal. I wanted one that is honest about progress and leaves every invitation and message to me.

What I built

1. 01

   Built the Next.js 16 App Router front end with React 19, Tailwind CSS 4, and shadcn/ui, including light and dark themes and lists that turn into cards on a phone.

2. 02

   Designed the Supabase Postgres schema with row-level security on every table, scoped to the owner, plus Auth (Google, password, emailed link), Storage, and pg_cron jobs.

3. 03

   Wrote safe-fetch, the only path for fetching outside URLs: it refuses private and reserved addresses, pins the connection to the addresses it just checked, and follows redirects by hand.

4. 04

   Built Discover, a research assistant over public sources that verifies each fact against the source page and drops any quote it cannot find, with scores computed by visible rules rather than a model.

5. 05

   Built the Jobs search over employers’ own boards, where a posting gets an Apply button only after its link is verified.

6. 06

   Wrote 1,523 unit tests and 305 pgTAP tests (checks that run inside the database itself), and ran axe-core accessibility scans in the Playwright end-to-end suite, with CI on GitHub Actions.

The code

/**
 * The follow-up touches after someone accepts an invitation, and a draft message for each. Drafts are built
 * from a template using only what the record holds (the person's name and, if known, their school and field),
 * never claim a conversation that did not happen, and are copied by the user, who sends them. Nothing here
 * calls an AI service, so a draft cannot contain an invented fact.
 */

export const FOLLOW_UP_KINDS = ["thank_you", "coffee_chat", "check_in_30", "check_in_90"] as const;
export type FollowUpKind = (typeof FOLLOW_UP_KINDS)[number];

export const FOLLOW_UP_LABELS: Record<FollowUpKind, { title: string; when: string }> = {
  thank_you: { title: "Thank them for connecting", when: "the day they accept" },
  coffee_chat: { title: "Ask for a short chat", when: "a week after" },
  check_in_30: { title: "Check in", when: "30 days after" },
  check_in_90: { title: "Catch up", when: "90 days after" },
};

Outcome

A deployed app on Vercel and Supabase that never scrapes LinkedIn, never sends anything for you, and shows progress that cannot be inflated by pending invitations.

What I learned

Security and honesty rules are easiest to keep when they live in one place: one function for outside fetches, one place for counting rules, and row-level security in the database instead of checks scattered through the app.

What I'd fix next

Run the screen-reader checklist (not done yet, and the README says so) and re-measure Lighthouse on production, which has not been repeated since the newest features landed.