I've hunted for a flat in Prague more than seven times, and every search was harder than it needed to be. PraguePad is my answer: a Telegram agent that watches Prague's rental market around the clock, scam-checks every match and alerts renters within minutes, with an AI-drafted first message in Czech and English. Designed, priced and directed by me, built with AI, and now in private testing while I run user research.
Renting in Prague is a race most people lose. Good flats go within hours, the quickest route to a landlord sits behind portal paywalls - the biggest charges 799 Kč just to send a message, and people still get ghosted - and community groups fill the gap with scams. I've lived this through more than seven searches of my own, and spent years watching, and helping, others struggle through the same thing. An unnecessarily hard process with an obvious fix, and nothing local was doing the whole job.
One bot, no app to download. Renters answer a short interview in Telegram - budget, districts, size, move-in date - and get an alert minutes after a matching flat appears across Prague's major portals and community sources. The same flat listed in three places arrives once.
Every alert scam-checked before it's sent. Each listing gets a green, amber or red verdict grounded in Czech rental law and known scam patterns. Amber arrives with a warning; red never reaches the renter.
Help applying, with a human in the loop. One tap drafts a warm first message in Czech and English. The renter edits it and sends it through the portal themselves. I chose not to auto-send: a bot messaging landlords on someone's behalf risks their account and raises legal questions the renter-sends route avoids.
Pricing for a product you're meant to leave. The plan started as a 199 Kč monthly subscription. I dropped it. A flat hunt works like a dating app - success is the moment you delete it - so a rolling subscription would charge people for the product working. PraguePad sells a one-off 2-month pass at 449 Kč with 48 hours free, priced against the time, effort and value behind it, and open to change once testers weigh in.
A free tier that's actually fair. The bot is public, so anyone can start it - which rules out a simple freemium switch. The hard part was a free version generous enough to genuinely raise someone's chances of a viewing, while leaving a real reason to pay. Free users get a daily summary of their best matches; pass holders get every alert live. One free trial per Telegram account.
Design: structure from AI, taste from me. The AI's first landing page had a useful structure and no taste. I designed the logo and visual identity, then redesigned the site mobile-first in Figma, because a Telegram bot is mostly reached from a phone. I checked that hunch against research: in Telemetr's 2025 Telegram audience study, 83% of respondents named their phone as the device they use Telegram on most, against 17% on a computer - one survey, not a global census, but a clear enough signal to prioritise the small screen. Claude built from my frames and I refined from there. Live in English and Czech at praguepad.cz.
How I worked. I leaned on AI heavily - Claude Code wrote and shipped the code. The product, strategy, pricing, free-vs-paid split and visual design were my calls, and I wrote each decision down as I made it. I tested every bot use case and workflow myself, and the build runs 732 automated checks. Next: user research and interviews with the first testers.