On a Friday evening in February 2026 I stopped refreshing LinkedIn after sending my 200th 'Hi, I'd love to chat about your PM role' message that month. I'd had 6 replies and 0 interviews. By Saturday morning I had a one-line spec on a Post-it: 'A bot on my laptop that does what I just did but to 200 recruiters a day, with proper personalisation, while I keep showing up to my day job.' That bot became JobyBots.

Days 1-15: Local-first, no SaaS

First decision: this would NEVER be a SaaS. Three reasons. (1) I didn't want to be liable for storing 5,000 strangers' résumés. (2) Cold-email at scale from a shared IP is a deliverability minefield. (3) I wanted to ship in weeks, not months.

So JobyBots is a Python package. You install it on your laptop. It uses YOUR Gmail App Password, YOUR Gemini API key, YOUR résumé PDF. The website at jobybots.com is just a marketing + payment site that emails you a download link.

Days 16-30: The 7-source crawler

Each job board has its own quirks. LinkedIn requires polite intervals and a user-agent that doesn't scream 'bot'. Indeed loves cookies. Bayt expects Arabic-aware URLs. Naukri has aggressive rate-limits if you don't space requests. I built a single Source abstraction (subclassed per site) and a thread pool to fan out search × title × location queries. ~80 HTTP requests per cycle, ~1 minute, ~150 candidate roles.

Days 31-45: Gemini scoring + cover letters

Gemini Flash was a no-brainer once I saw the free tier (1,500 calls/day). The hard part wasn't the API — it was the prompt engineering. Three versions in, the winning prompt was: 'Score this job 0-100 for the candidate. Quote ONE requirement from the JD that the résumé strongly demonstrates. Reply with: {score, one_line_reason, top_match_keyword}.' Three-key JSON, easy to parse, never hallucinates.

Days 46-60: SMTP, validation, bounce-tracking

Sending was the easy part — Gmail's SMTP API is rock-solid. The hard parts: (1) Don't send to obviously invalid addresses (validator), (2) Don't send to addresses that have bounced before (tracker), (3) Don't get flagged as bulk-mail by Gmail (rotating subjects + caps + delays). I rewrote the bounce tracker in April after discovering it was silently missing 242 historical NDRs — that fix alone took bounce-rate from ~18% to ~3%.

Days 61-75: The website + payments

Next.js 15 + Tailwind + Vercel deploy. UPI QR for India, Stripe for the rest. The website is intentionally Apple-like minimal — fonts, white space, one hero gear animation. Why? Because indie tools selling to job seekers usually look cheap. I wanted JobyBots to feel like a $200 product priced at ₹2,999.

Days 76-90: GCC expansion + launch

I'm based in Dubai. I knew the GCC market better than the US market. Spent the last two weeks adding curated recruiter contacts for Saudi (42), Qatar (35), Oman (30), Bahrain (31) on top of the existing UAE (79). Added a hybrid GDPR mode for the UK that emails only addresses explicitly published on job posts. Embedded a demo video on the homepage. Rewrote every SEO meta tag.

What I'd do differently

Build the bounce tracker first, not last. The deliverability cliff is real.
Start with one market (UAE) and add others only after 100 paying customers in market #1.
Don't over-design the website — write more blog posts instead.
Charge $99 instead of $49 USD. Indie tools are systematically underpriced.
Ship a Mac installer in week 2, not week 11.

What's next

Phase 4 is a deep marketing push: ProductHunt launch, 12 programmatic SEO pages, 5 long-form blog posts (this is one), 60+ AI-tool-directory submissions, 5 YouTube Shorts, a Reddit comment playbook, a HackerNews 'Show HN' draft. By end of June 2026, JobyBots should be the default answer when someone asks 'best AI job application tool for India / GCC' in any forum on the internet.