Skip to content

Latest commit

 

History

History
98 lines (73 loc) · 4.22 KB

File metadata and controls

98 lines (73 loc) · 4.22 KB

JobSearchBot — Vision and Startup Idea

This document holds the project's purpose, full feature set, and phases, so nothing is forgotten. Last updated: 2026-07-14


1. Purpose

An AI agent that, on the user's behalf, scans 24/7, understands, selects, and (if allowed) applies.

Problem: People are job hunting but cannot check every site 24/7. There are many listings; some people can't see them all, others can't be bothered, and they miss opportunities.

Solution: While the user goes about their own work, the bot watches all active listings in Azerbaijan for them, finds the ones matching their profile, notifies them, and — if permitted — applies.

Opportunity: There is no local competitor doing this in Azerbaijan.

2. Who it serves

  • Job seekers (Azerbaijan) → the user-acquisition engine (cheap/free)
  • Employers → the real revenue source (Phase 3)

3. Product and features (full list)

Phase 1 — Multi-user smart notifications (free)

  • The user signs in and uploads a CV
  • Email association
  • Manual filters: job type, working hours, salary, etc.
  • or setting up filters by talking to the AI in chat inside the app
  • or fully autonomous mode: the AI reads the CV and decides for itself what fits
  • All active listings in Azerbaijan are scanned (multi-source: jobsearch.az + similar sites)
  • The user chooses the notification frequency (e.g., every 2-3 hours)
  • Every listing includes the application deadline

Phase 2 — Automatic applications

  • Trust ladder: notify only → one-tap confirmation → fully autonomous
  • Automatic email to every suitable job
  • A job-specific cover/motivation letter (written by the AI)
  • Sends only the CV if only a CV is requested, or written text if text is requested
  • Fills in web forms if the site requires it
  • Letter-source option: the user writes their own in advance / a local LLM / our AI / an external API (the user approves)
  • Applications go out from the user's own account (no detectable pattern)

Phase 3 — CV Pool (CV House)

  • The user places their CV into the pool of their own will and consent (like an ad)
  • Companies look at the pool when searching for suitable candidates
  • The AI selects/ranks the most suitable candidates
  • Employers pay → the real revenue

4. Business model

Side Price Role
Job seeker cheap/free (e.g., 5 AZN/month) growth engine
Employer the real fee (candidate access / subscription) revenue

This is essentially the Indeed/LinkedIn model: free for job seekers, employers pay.

5. The 2 metrics to measure (everything else is noise)

  1. Phase 1 — retention: is the user still using it after 3-7 days?
  2. Phase 2 — response rate: does auto-apply actually produce interviews/responses?

6. Risks (don't forget)

  • Success = churn. A user finds a job in a month and leaves → lifetime value ≈ 1 month. Growth happens only through a constant stream of new users → design for virality/word of mouth.
  • Two-sided market (Phase 3): employers won't pay for an empty pool → win the job-seeker side first.
  • CV = personal data: opt-in is mandatory; a leak = a trust and legal disaster.
  • LLM cost can blow up the unit economics → the user's own key / a local LLM.
  • AI-native alone is not a moat. The real moat: the CV pool + a data flywheel that learns from hiring outcomes.
  • Platform/ToS risk: sites may block us → multiple sources reduce this risk.

7. Positioning: AI-native

Remove the AI and the product collapses (understanding CVs, comprehending listings, writing letters, autonomous applications). Competitors are a keyword database; we are an agent. The current code is still rule-based (regex); the move to the AI-native core happens after real users arrive.

8. Current state

A working single-user Telegram bot:

  • Runs 5 times a day on GitHub Actions (Baku 09/12/15/18/21), free, on nobody's computer
  • Scans all ~2700 listings on jobsearch.az (API + pagination + retry)
  • Matches against the CV profile and sends a single batched message to Telegram (with application deadlines)

Next step: Phase 1 — make the bot multi-user.