01 Whoami

sshlg@me ~ % whoami

Sergey Sheleg

Technical entrepreneur · products used by millions

Technical entrepreneur in Warsaw, Poland. 13 years building and launching products used by millions of people, most of that on mobile. I run the whole loop on a product: research it, build it with agents, take it to market, and know what it costs to run every month (that last step is the one most builders skip, and it's the one that decides whether the thing survives its first year). Then I run it again. The way I do it is open source: 23 skills any coding agent installs in one command, written down in the first place because a process I can't hand to somebody else is a habit rather than a method. Same loop, every time.

Mobile products
Agent systems
Growth & distribution
Go-to-market
AI cost engineering
Monetization
Market research

Work with me What's running contact@sshlg.me

Years: 13+ shipping
Shipped: 10 products
Live: 6 running
Toolkit: 23 skills, MIT

02 Toolkit

13 years, installable.

Left alone, a coding agent produces volume: plausible code nobody can ship, scale or maintain. So I wrote down the way I actually take an idea to production (discovery, gated delivery, design, copy, search, payments, agent orchestration) and packaged it as skills an agent follows instead of improvising, which is the difference between a process and a preference.

I use this set every day, on the products below and on the page you are reading right now. It grows whenever the work teaches me something (the version numbers keep moving for that reason, not because anything is being rebranded). What does it cost? Nothing. It's MIT, it carries no telemetry, and it has never been for sale.

Packs: 8
Skills: 23
Agents: 70+
License: MIT

install $ npx sshlg-skills install

One command, every agent · MIT, no telemetry · sshlg-skills · skills.sh/ssheleg

6 slash commands · Claude Code as plugins, plus Cursor, Codex, OpenCode, Gemini CLI, Windsurf, Zed and the rest of the 70+ agents the skills CLI supports.

03 Projects

What's running.

Five products, ordered by how many people they reach rather than by how recently I started them (the recency order is the flattering one, and it would have put a product with a hundred users above one with fifty-six million installs). Two were co-founded with real teams around them and three I build and run through SV Lab. Everything else I have shipped sits in the ledger further down, with the role I actually held on each one rather than the role that reads best.

Co-founded, at scale

Nicegram LIVE
nicegram.me
56M+ organic installs

A Telegram client for the people already living in that ecosystem, and the largest thing I have co-founded.

Five years from a hypothesis about selling info-products inside Telegram to a client people find and install on their own, which is a longer road than the number suggests

SEO, referral and viral mechanics, and product work. No paid acquisition, which is also why the running cost looks the way it does

GPL-2.0 on all three platforms, so the privacy claim is something you can read rather than something I assert

Co-founder · growth

TelegramGPL-2.0Open sourceGrowth iOS Android Desktop
eSIM Plus LIVE
esimplus.me
Two products sharing one app: travel data over eSIM, and virtual numbers that receive the SMS a foreign service insists on sending.

On the App Store since December 2019 and still shipping (longer than most of this category has existed, travel eSIMs included)

Regional and global plans, so one eSIM can cross a continent without a swap at the border

The numbers half is what makes it a second line rather than a travel SIM, and it earns twice over, once from the traveller who needs data and once from the person who only ever needed somewhere for a verification code to land

Co-founder · product

eSIMiOSAndroidWebPayments App Store

Building now, under SV Lab

Prowl LIVE
prowl.chat
One MCP endpoint, and 448 market-intelligence tools behind it.

448 tools across 17 providers on one endpoint, covering SEO, ads, SERP, reviews, pricing and trends, so an agent asking a competitive question does not need seventeen API keys and seventeen billing relationships to answer it

33 modules in the catalogue, up to 16 in a single report, delivered as a deck, a PDF, an infographic or an audio summary

Pay per call. A deep report runs about $8.60 in credits and the free tier opens with $5, so the first one costs nothing to try

Founder · build · GTM

MCPTypeScriptNext.jsLLM orchestrationStripe
PrivateClawd LIVE
privateclawd.com
An AI agent on a machine of its own, running in 60 seconds.

60 seconds to a running agent on a dedicated VM, against roughly three hours to stand the same thing up yourself

Token caps, loop detection and auto-pause, because the failure mode of an agent nobody is watching isn't a crash you notice on Monday, it's an invoice you notice at the end of the month

From $30 a month for one agent and $20 each past ten, with prompt-injection and link-preview exfiltration defence in the box

Co-founder · build · GTM

Cloud orchestrationDockerOpenClawStripeEdge
CheckMyData LIVE
checkmydata.ai
Ask a database a question in English, and get back the SQL that was actually right.

PostgreSQL, MySQL, ClickHouse and MongoDB, read-only by default

It indexes the codebase as well as the schema, so a question about revenue resolves against the business logic rather than against a column that happens to be called revenue

MIT, self-hostable on Docker, no telemetry. Nothing about the data leaves the machine it runs on

Founder · build · GTM

TypeScriptPostgresDockerLLM agentsOpen source

The whole portfolio

10 shipped · 6 still running · 3 sold · since 2013

Every product I have shipped or co-founded, newest first, with the role I held on it. What is deliberately missing? Employment inside somebody else's company, which lives in the biography below instead, because this is a list of things I owned a piece of and a job is not one of them (the Ultimate Guitar years were the largest audience I have ever worked on and they still do not belong in a ledger of ownership).

04 Path

How I got here.

13 years of shipping, plus the years that led into it. The degree pairs engineering with economics, which sounded like a bureaucratic compromise at the time (nobody wants a diploma with three professions hyphenated into it) and turned out to describe the whole job: the code and the pricing are the same problem seen twice, and anyone who only sees one of the two builds the other one badly.

Before 2013 Mogilev · Minsk

Olympiads, then a degree that named the job.

Started programming as a child in Mogilev and won school informatics olympiads. Graduated from BSUIR as an engineer-programmer-economist, a title that reads like a committee assembled it and turned out to describe the job exactly: the code and the business model are the same problem seen twice.
2013 → 2015 Elinext · HiQo Solutions

Other people's products, other people's deadlines.

Elinext from 2013, HiQo Solutions from 2014. Software to order for European and American clients, then products for companies in tech, agro, AI and hardware. Enterprise delivery, one release after another, on somebody else's clock (which is a different clock from your own in exactly the way that matters, because you cannot negotiate with it). This is where shipping software stopped being the same skill as writing it, and that is a distinction you only really learn by missing a date that belonged to someone else.
2015 → 2020 Muse (Ultimate Guitar)

The world's largest musician platform.

Head of Android for the Ultimate Guitar app, and growth product for a platform with 350M+ users. I launched MelodiQ, which recognises music by ear and turns it into something to practise, and built the cross-platform engine that renders and plays back scores (it lives in MuseScore now, the open-source notation editor, which is a longer afterlife than most code gets). Someone else's company, and still the largest audience anything on this page has reached.
2019 → now Own products

Shipping for myself.

Started shipping products of my own: consumer apps, connectivity, music, messaging, and now AI. 10 in the ledger, 6 still running, 3 sold and still being updated by whoever bought them. Co-founder on 8 of them, founder of 2. The first two shipped while the Muse job was still running, which is what starting out actually looks like from the inside.
2021 → now Nicegram

Five years inside Telegram.

It started as a hypothesis about Telegram as a sales channel for info-products, which is not what it became. Several pivots later it was Nicegram, a client for the people already living in that ecosystem. 56M+ organic installs in five years, out of SEO, referral and viral mechanics and product work, with no paid acquisition behind any of it. Nicegram Business is the next turn of the same track, in early access at nicegram.ai.
Now Warsaw, Poland

AI products, and the toolkit under them.

Building AI products under SV Lab and inside the Nicegram platform, and packaging the way I build them as open-source skills any coding agent can install. Why publish the method at all? Because a claim about how somebody works stays unfalsifiable until the method is installable, and once it is installable anyone who doubts it can run the thing against their own repository and watch it either hold up or fall over. That is the whole argument.

What it proved.

Where the 13 years went, and the artefact standing behind each row. The bar is relative weight across my own work rather than a score against anybody else's (a scale that compared me to the field would need a field to compare against, and there isn't one), and a row that can't name something you could go and check doesn't appear on the chart at all.

Mobile products Head of Android at Ultimate Guitar · 3 iOS products built and sold · Nicegram on iOS, Android, desktop
Growth & distribution Nicegram · 56M+ organic installs, no paid acquisition
AI cost engineering My own agent spend · $10k+/mo → ~$1k
AI agents & orchestration Prowl · 448 tools behind one MCP endpoint
Entrepreneurship 13 years · 10 shipped · 3 sold · co-founder on 8
Product vision Nicegram Business · a paid layer over an app that never had seats
Delivery & project management task-pipeline · ten gated stages, brief to acceptance
Market & competitor research Prowl · 33 modules: SEO, ads, funnels, reviews, pricing
Consumer product at scale Muse (Ultimate Guitar) · Head of Android and growth product, 350M+ users
Monetization Subscriptions, pay-per-call wallets, crypto rails
Funnels & conversion eSIM Plus · paid acquisition to activated traveller
Open source Nicegram GPL-2.0 · CheckMyData MIT · 8 skill packs MIT
Infrastructure PrivateClawd · isolated VM per customer, 60-second deploy

05 Hire

Two ways to work with me.

Prices sit on the page rather than behind a call. Why print them? Because the number decides most of these conversations anyway, and discovering it on a call costs us both half an hour to learn something a line of text could have said in advance (the half hour is mine as well as yours, which is the part that makes this self-interested rather than generous). Both engagements start the same way: an email that names what you want and by when.

Build

$5,000–10,000 fixed, agreed before we start

2 weeks from brief to deployed

Your hypothesis, live in two weeks.

Bring one idea. Two weeks later it's a working product on a real domain, deployed, and yours (the domain and the accounts included, which is the part that decides whether you own the thing or only rent it from me). Not a prototype, not a Figma file, not a deck.
- Scope locked on day one, because the brief is the contract and a moving brief is how two weeks becomes two months
- Product, landing page, payments and analytics wired to each other rather than shipped as four separate things
- Deployed on your infrastructure, under your accounts
- Source, docs, and the agent toolkit that built it, handed over at the end
Who it's for

Founders and teams with one hypothesis worth testing and no appetite for a three-month build.

The limit

One hypothesis per engagement. Where two weeks isn't enough, I say so before taking money rather than after, and I'll usually say what two weeks does cover instead.

Send the brief
Training

$5,000 per session, paid up front

2 hours live, one session

Stop overpaying for AI development.

Two hours, screen to screen, on the setup that took my own agent spend from over $10,000 a month to about $1,000. What to change, in what order, and what the current way is costing while you decide (the run rate keeps running during the conversation, which is the only urgency here that is real).
- Where the spend actually goes, measured on the real usage rather than guessed from the invoice total
- Model routing, context discipline and caching that survive real work rather than a demo
- The skills and hooks that stop an agent burning tokens on the wrong task, which is where most of my own bill went
- The before-and-after numbers, so the result is something you can check without me
Who it's for

Solo builders and teams running Claude Code or a comparable agent daily, with a bill that has stopped being funny.

The limit

The 10× is my result on my stack, not a guarantee on yours. What I can promise is the method and a measurement at both ends.

Ask about a session

Anything that fits neither is still an email. An audit, a longer build, a second pair of eyes on an agent stack that has started costing more than it returns, or a question about whether the thing is worth building at all (that last one is free, and it has talked more than one person out of a project they were about to pay someone for). I answer either way, and where the honest answer is that somebody else is a better fit for the job, that is the answer you get.

06 Writing

Two channels.

I post about what I'm building while it is still in progress, in Russian and in English. Same subjects, different rooms, and neither channel is a translation of the other (a post gets written in the language it was thought in, which is why the two don't line up week to week).

The longer write-ups land on the blog, where every checkable sentence carries its proof at the bottom of the page.

07 Contact

Reach me.

Email works best. Tell me what you want and by when, and I'll tell you straight away whether it fits (a two-line brief is enough to answer that, and a deck rarely is). Terms for the two paid engagements are above, and anything that isn't one of them arrives at the same address anyway.

Sergey Sheleg

13 years, installable.

What's running.

The whole portfolio

How I got here.

Olympiads, then a degree that named the job.

Other people's products, other people's deadlines.

The world's largest musician platform.

Shipping for myself.

Five years inside Telegram.

AI products, and the toolkit under them.

What it proved.

Two ways to work with me.

Your hypothesis, live in two weeks.

Stop overpaying for AI development.

Two channels.

Reach me.