Anton Kalashnikau · Side projects

Things I built to find out.

The day job is leading a team of 7–10 engineers, and that is the point. These are side bets, each sized to answer one question, not to become a company. The discipline is the one I bring to a roadmap: decide what not to build, and when to stop.

For each one: the question, the smallest thing that could answer it, and the answer.

  1. 01

    AvocadoScore

    Live, free to run

  2. 02

    dachzgory.pl

    Pilot with roofing firms

  3. 03

    Offramp

    Live

01 — Live, free to run

AvocadoScore

A website audit that scores a site for Google and for AI crawlers in one report, with the fix list first.

Stack
Next.js · TypeScript monorepo · PageSpeed API · Stripe
The question
AI assistants now fetch pages the way search engines do, and most sites have never checked what those crawlers see. Could one report score a site for both, and say what to fix first?
What I built
A scanner with a 12-point SEO scorecard, an AI-readiness check (llms.txt, crawler policies, structured data, server-rendered content), Core Web Vitals, a prioritised fix list that exports to Jira and Linear, a public index of about nine hundred audited sites, free tools, a CLI and payments.
The answer
Yes, and the score turned out to be the least useful part. What people act on is the fix list under it, so that is where the work went: each finding specific enough to hand to an engineer as a ticket. It stays live as a free audit.
Checks per audit
20+
Sites in the public index
~900
Pages per full scan
500

02 — Pilot with roofing firms

dachzgory.pl

Type a Polish address, get the roof measured from open geodata in seconds. No ladder, no drone.

Stack
Python · FastAPI · DuckDB · LiDAR · GUGiK open data
The question
A roofer has to drive out and measure before they can quote. Poland publishes 3D building models, a national elevation grid and LiDAR, all free. How far does open data get before someone has to climb a ladder?
What I built
An engine that finds the building from an address, rebuilds the roof from the LoD2 model where one exists and from the elevation model or LiDAR where it doesn’t, and returns surface area, pitch, ridge, valley and eave lengths, with a PDF and a bill of materials a roofer can hand to a client. Every county in Poland is covered one way or the other. Pilot firms get a code, measure their own roofs and mark each result right or wrong.
The answer
Checked against the official 3D models on roofs that hadn’t changed, area and pitch come out close enough to quote from. Against real tender bills the plan area lands within about a tenth. Good enough to quote from, not yet to order materials from; that gap is what the pilot is for.
Area error, median
0.7%
Pitch within ±2°
94%
Counties covered
380 / 380

03 — Live

Offramp

A cloud exit calculator: one month of an AWS, Google Cloud, Azure or Oracle bill, priced on Hetzner, OVHcloud, Scaleway or DigitalOcean, net of the engineers it takes to run it.

Stack
Next.js · DuckDB-WASM · TypeScript, unit-tested estimator
The question
Same domain as the day job, opposite question. By day I help teams spend less inside Kubernetes; the louder question this year is whether to be in the cloud at all. The honest answer needs your bill, not someone else’s blog post. Could a tool give it in a minute, without the bill ever leaving your laptop?
What I built
A calculator that reads AWS, Google Cloud, Azure and Oracle cost exports inside the browser (DuckDB compiled to WebAssembly; nothing is uploaded), maps every service to its replacement on Hetzner, OVHcloud, Scaleway or DigitalOcean, grades each one Easy, Moderate or Hard, and takes off engineer time, on-call and the migration one-off before it calls a saving a saving. Every rule, price and ratio is published on the method page with its source.
The answer
The number people want is the saving. The number that decides is the engineer-time line underneath it, and most cloud-exit stories leave it out. Live and open about its model; what it needs next is real bills from teams who are actually deciding.
Clouds it reads
4
Destinations priced
9
Uploaded to a server
0 bytes

Let’s talk.

Have a question of your own that a few weeks of building could answer? Twenty minutes, no slides.

Book a callkalashnikau@gmail.com