Athens, Greece · 17
Nikolas Sapalidis
I build custom software — websites, apps, AI systems — from scratch.
I started where most teenagers start: deep in the self-improvement rabbit holeBooks, systems, 5am routines. Useful discipline, zero income.. That led to trading and crypto, then setting up a Google Ads agencyBuilt the offer and the site, never launched it. No clients, no ads ever run. that never launched, then trying to sell digital products — and getting nowhere. Didn't make a cent from any of it.
Then I learned to code. Got my first paying client in February 2026 and realized it was actually working. Since then I've built across the full stack — websites, web apps, iOS apps, AI automations, custom pipelines, admin dashboards, internal tools. No templates, no page builders. Everything custom.
Alongside client work I build products. First was Helpmarq — a feedback marketplace for structured, expert feedback on real projects. Since then: MarketMyAppMarketing plan generator for indie app builders — positioning, channels, copy., Creator RoastHonest teardown of a creator's page. No compliments, just what's costing you signups., Branch AI (900+ npm downloads), NeuroPulseOpen-source brain-response analyzer for marketing content. Drop in an ad, get scored across 8 brain regions., ns-ui (298 React components you install by URL), and more. I also work at AutomateSphere AI building custom AI automation for businesses.
"No existing brief. Needed something premium without feeling generic. Four days later: live, custom, exactly what I asked for."
— Sean Lee, client
Since April 2026 I've been on Claude Code MaxAnthropic's top tier. Custom skills, hooks, subagents — the whole harness tuned around how I actually work.. It changed how I work. I run a homelabSelf-hosted Supabase + n8n on my own hardware. No per-seat pricing, no vendor deciding my rate limits., ship multiple products in parallel, and build client work on the side. I write about the patterns that actually work — not the hype.
Lately I've been going deep on machine learningBackprop by hand, tensors from scratch, small models I can explain end to end. No framework hiding the maths. — from the fundamentals up, not the API-wrapper version. I've shipped a lot of AI on top of other people's models; now I want to understand what's actually happening underneath. Building small models by hand, doing the math, no shortcuts.
1984 contributions this year
Currently
- → SAC CapitalAgent that screens equities on fundamentals, momentum and disclosure signals, then argues its own picks down. Real money, published CLI. — live AI trading agent, CLI on PyPI
- → Learning ML from the fundamentals
- → ns-ui298 React components, install by URL. Playwright gate fails anything whose hover looks identical to its resting state. — component registry, live at design.helpmarq.com
- → sigevalEvals as proportions with confidence intervals. PASS, FAIL, or INCONCLUSIVE — never a coin-flip. — statistical evals for LLMs
- → AutomateSphere AI — client automations
- → Client dev work — web, automations
How it works
- 1. Blueprint call — 30 min, you describe the problem
- 2. I send a scoped proposal within 24 hours
- 3. We build — Figma first, code after approval
Contact for more.
Taking 1–2 clients this month — slots fill fast.
Build something
Free AI System Blueprint Call
30 min. No pitch. Just a blueprint.
Here for the content
Claude Code Starter Pack
Skills, hooks, CLAUDE.md template. Free.
Not ready yet? Get the build log — weekly notes on what I'm shipping, patterns that work, things that didn't.