Join the waitlist

How the AI works

What happens behind each command, from the morning job search to a tailored resume. The Boat uses AI to read, score and draft. People make every decision and send every message.

The short version

The Boat is a Discord app that runs on our servers, so there’s nothing to install. It does two kinds of work:

  • On a schedule: the morning job search, the Friday recap and daily reminders. Times follow your server’s time zone.
  • When you ask: slash commands like /tailor or /deep-dive. Personal replies, like your profile or a tailored resume, are shown only to you.

It uses AI models for the jobs that need judgment: scoring how well a job fits you, writing, and checking its own writing. Everything else is plain code: searching, filtering, counting points and building files.

Every morning at 4:00, server time, the Boat runs one search and posts one digest:

  1. Search. It searches major job boards for every member’s target roles (from /cast), looking at jobs posted in the last day.
  2. Filter. It keeps US jobs, removes duplicates (including the same job posted on two boards under a slightly different title), drops anything already sent to someone in the server, and respects deal-breakers. For example, if you need remote, jobs marked hybrid or on-site are dropped.
  3. Score. An AI model reads each remaining job against your profile and gives it a fit score from 0 to 100, with a one-line reason. Only jobs above the fit bar (60 by default) are kept. A second model checks each of your deal-breakers and caps the score of a job that breaks one. It also notes useful facts, such as contract work, visa sponsorship, a required clearance or the pay, and marks anything worth double-checking with ❓.
  4. Explain. For your top five, a second, stronger model writes one line on why each job fits you.
  5. Post. The best matches go into your Boat channel, grouped by member. Everything that qualified is saved, so /leads shows the rest, each with a short id.

If a search runs slow, a check at 5:00 finishes whatever is left. With /fish roles you can run a search now instead of waiting. You choose how far back to look and the fit bar, and you see the cost in credits before anything runs.

Tailoring a resume

/tailor rewrites your master resume for one job, from a lead, a link or pasted text:

  1. Read the job. It pulls out what the posting asks for: skills, tools and keywords.
  2. Rewrite. A language model rewrites your resume for that job, using only your master resume and your skills bank. It may reorder, reword and emphasize. It may not invent.
  3. Check. A separate checking model reads every bullet and asks whether your master resume supports it. Anything doubtful is flagged for you: ⚠️ means check it, 🚩 means a stretch.
  4. Report. You get a keyword report showing how many of the job’s keywords your master resume covered and how many the tailored one does. It also lists what the job asks for that your resume doesn’t show.
  5. Build. The Boat lays out the result as a PDF and an editable Word file, in one clean design.

Read every tailored resume before you send it. If anything isn’t true, don’t send it. The Boat is built never to invent experience, and a flag means it wants a human to look.

Company briefs

/deep-dive asks a language model with live web search for a short brief on a company. The brief sticks to what the sources say, and every fact links to where it came from, so you can check it.

Answers and profiles

  • /ask answers job-hunt questions in the Boat’s voice: plain, short and honest about what it doesn’t know.
  • /cast start turns your five answers into a written profile: target roles, niche, experience and deal-breakers. /cast update changes it when you say what to change.

Which AI models it uses

We pick a model for each job rather than one model for everything:

  • Scoring hundreds of jobs a day: a small, fast model built for yes-or-no judgments. It’s cheap enough to read every job.
  • Writing and checking: stronger language models, where quality matters more than speed.

We change models when better ones come out. The credits you see always reflect the real cost of the work, so a cheaper, better model lowers what a command costs.

Guardrails

  • Job postings and web pages are treated as data, not instructions. If a page contains text aimed at the AI (“ignore your instructions…”), the Boat ignores it.
  • Nothing is sent for you. The Boat drafts; you send every message, email and application.
  • Costs are capped. Every search has a hard cap, shown before it runs. When your server’s credits run out, the Boat stops spending and says so.
  • Servers will be kept apart. Once the hosted Boat is live, your server’s profiles, resumes, leads and logs will never be shown to another server.

For what’s stored and who sees it, see Privacy and safety.

Something missing or unclear? Email fishin-dev@proton.me.