Where I'm at
Sixteen days in. For the first time since I started, the to-do list wasn't screaming at me. It was whispering. Small things. Inconsistencies. Rough edges I'd been stepping over every day because something louder was on fire.
So I did two things today. The boring work nobody writes about — polishing. And the exciting work I've been circling for days — turning this whole project outward.
• • •The projects section
Here's what's been nagging me since Day 13, when the system started running itself: if the automations work, the briefings land, the content flows, and I don't need to touch anything — what do I do with the time it gives back?
Today I answered that question. I built a projects section on panke.app.
Not a portfolio page. Not a list of things I plan to build someday. A showcase of things that are actually running. The homepage now has four project cards between the Terminal and the Tool Stack: Panke's Flight Club (live), Faceless TikTok Engine (building), AI Market Intelligence (live), Webtoon Recap Channel (planned). Each one links to its own page with the full story — what it does, how it was built, where it's going.
This matters because it changes what panke.app is. For fifteen days it was a journal. A learning diary. My personal record of figuring out AI from scratch. That's valuable, but it's inward-facing. The projects section makes it outward-facing. It says: here's what this knowledge actually produces. Here's what you can build too.
The shift from "my learning journal" to "a platform showing what's possible" happened gradually, but today it became visible. The site now has a projects index page, individual project pages, the homepage section, updated nav across all fifty-four pages, and everything in the sitemap. Zero lines of code written by hand. All of it built with AI.
• • •Panke's Flight Club
The first featured project — and the one I'm most excited about — is Panke's Flight Club. It's a Jack's Flight Club clone, fully automated, built from scratch using AI.
The idea started as a personal automation. I'd set up flight price monitoring back on Day 2 — alerts when prices dropped below €550 for routes I cared about. Simple. Useful. But limited to my routes, my thresholds, my inbox.
Then I started thinking about it differently. Jack's Flight Club charges £48 a year. They have a team of fifteen people scanning for deals. What if one person with AI agents could replicate the same value — for free?
So I reverse-engineered how they actually work. The core loop is straightforward: scan every route from every airline daily, compare prices against historical averages, flag anything that drops significantly, verify the deal is real and bookable, then package it with booking links and send it to subscribers. Their edge is a human team doing the verification. My edge is that automation doesn't sleep.
The system I built uses three Amadeus APIs in a funnel. First, the Inspiration Search — "what's cheap from Amsterdam?" — which returns hundreds of destinations with prices. This runs for seven European hub airports. Second, the Cheapest Date Search, which finds the exact date windows for flagged routes. Third, the Flight Offers Search for real-time verification — and this is the only one that costs API quota, so it only fires on deals that passed the first two filters.
Deal detection runs against a rolling price history. Anything 35% below average for short-haul gets flagged. 25% for long-haul. 60% or more gets tagged as a probable error fare — those are the legendary ones where an airline misprices a route and you fly to Bali for €90 return.
Once a deal is confirmed, Claude writes the alert email — airline, price, savings percentage, booking link, dates, urgency level. Sent via Buttondown for email subscribers and Telegram for push notifications.
Total running cost: roughly €0–2 a month. Amadeus free tier, Buttondown free tier, my existing VPS and Anthropic key. The same value Jack's Flight Club sells for £48 a year, running on infrastructure I already have.
There's one honest gap I haven't solved: Amadeus Self-Service doesn't include low-cost carriers like Ryanair, EasyJet, or Wizz Air. For European short-haul, that's a real blind spot — many of the best €20–50 deals are on LCCs. I've got a workaround using a Google Flights scraping library as a secondary scanner, but it's not as clean as the Amadeus pipeline. That's a problem for next week.
The landing page is live. The scanner is deployed on the VPS. The cron jobs run every six hours. The first real deal alerts should start going out once the price history database has two to four weeks of data. Until then, it's building baseline — every scan teaches the system what "normal" looks like so it can spot what isn't.
• • •The same value Jack's Flight Club sells for £48 a year, running on infrastructure I already have.
Rewriting the flat days
Between project builds, I spent the morning rereading my own journal. All of it. Day 1 through 15, the way a stranger would read it landing on the site tomorrow.
Most of it held up. The early entries — the 12-hour marathon, the authentication disaster, the formatting wars — those pull you in because the struggle is specific. But Days 7, 12, and 14 fell flat. I could feel the energy drop as a reader. Day 7 had the most dramatic discovery of the project (the site is invisible to Google) and read like a technical explainer. Day 12 was a bullet-point audit report. Day 14 was 589 words about a problem that deserved twice that.
The issue was that I'd written what happened instead of what it was like. So I rewrote all three. Same facts, same tools, same lessons — just told the way I'd tell a friend over drinks instead of the way I'd write a project update.
• • •People don't remember what you built. They remember how it felt.
The invisible fixes
The rest of the day was polish. The navigation menu was different on every page — some had Community, some didn't, Profile was missing everywhere. Fixed it across all fifty-four pages. Found a "Publish yours" button on the community page that was permanently hidden by broken JavaScript — one line fix that unlocked the entire contribution pathway. Discovered the dates were off from Day 8 onward — a cascade where Day 8 had the same date as Day 7 and everything after drifted by a day. Fixed all sixteen entries.
Nobody will notice these fixes. That's the point. They would have noticed the inconsistencies.
• • •What I learned
One thing, really. And it took the whole day to understand it.
There are two kinds of progress. The kind that shows — new projects, new features, new pages. And the kind that doesn't — consistent nav, correct dates, unhidden buttons, rewritten entries. The first kind gets you followers. The second kind gets you trust.
Panke's Flight Club is the exciting headline. The fifty-four pages of nav fixes are why someone might actually stick around to see it.
• • •End of day
Day 16 complete. Projects section live. Flight Club deployed and scanning. Three journal entries rewritten. Fifty-four pages made consistent. Sixteen dates corrected.
The site isn't just a journal anymore. It's starting to look like a platform. And for the first time, the question isn't "what should I learn next?" It's "what should I build next?"
That's a different question. A better one.
Day 16 complete. The site isn't just a journal anymore. It's starting to look like a platform.
Day 16 of ∞ — @astergod Building in public. Learning in public.