Open for new engagements

I help SMB operators ship AI & automation that actually move revenue.

Operator at two companies. I run the same playbooks I advise on — outbound, signal pipelines, internal tooling — and only recommend what I've shipped against my own P&L.

WhatsApp Telegram
Selected work · 2024–2026

Six systems, shipped & running.

Each one I built, deployed, and operate today — the wins and the wrong turns.
01 — EDITORIAL AUTOMATION

An autonomous newsroom — 24 articles a day, zero human in the loop

Headless CMS + Python publisher. Source router, enrichment, image rehosting, deduplication. Newsroom-grade output running on cron in a regulated vertical.

Python · Headless CMS · GPT-4o · cron
RESULTS
720+articles / month
0human edits / piece
99.4%pipeline uptime
Read full case study →
$ tail -f publisher.log[10:14] router → 31 candidates[10:15] enriched → 22 ok[10:18] published → 22 ✓avg latency 4m 12s · uptime 99.4%
02 — EDITORIAL AUTOMATION

A Telegram-in, newsroom-out editorial bot

Drop a link or a raw post into a Telegram chat. The bot enriches it, drafts the full piece in the newsroom's voice, and publishes it to the headless CMS — minutes from tip to byline.

Telegram Bot API · GPT-4o · Headless CMS
RESULTS
4 mintip-to-byline
60+pieces from drops
0CMS clicks / piece
Read full case study →
YOU · 09:47https://example.gov/press-release/2026-q3-procurement-tranche
BOT · 09:51 · DRAFTED ✓720 words · published to /news/eu-procurement-tranche · 4m 12s end-to-end
03 — INTERNAL INFRASTRUCTURE

A private semantic-memory layer across every project I run

Vector DB + FastAPI + MCP behind a private network. Syncs project context across three machines. The substrate that lets one operator carry the workload of a small team.

Vector DB · FastAPI · MCP · Private mesh
RESULTS
14projects indexed
38kembeddings
<2scontext recall
Read full case study →
14
PROJECTS INDEXED
38k
EMBEDDINGS
3
DEVICES SYNCED
private endpoint · last sync 00:02:14
04 — WHAT I LEARNED

A productized outbound stack — 663 sends, 0 replies, one expensive truth

Sequence engine, reply detector, dashboard. Dogfooded it. The system worked; the offer didn't. Shelved the product, kept the lesson: validate the wedge before you scale the infrastructure.

n8n · FastAPI · Brevo · Postgres Read full case study →
The system worked end-to-end. The market told me the offer was wrong. That's the cheapest lesson outbound has ever taught me.— Postmortem, archived 2026-04-15
05 — WHAT I LEARNED

An automated trading bot — and the liquidity lesson that ended the thesis

Daily up/down markets on a major prediction venue. News + sentiment signals, paper-trade gates, full settlement loop. The strategy worked on backtest. Live spreads ate the edge before the first real position closed. Killed the bot before it killed capital.

Python · CLOB API · WebSockets · cron Read full case study →
PAPER P&L+18.4%
LIVE P&L−3.1%
SPREAD COST21.5%
The gap between a thesis and a market.
06 — CONSUMER PRODUCT

Bedtime Stories — a sibling-aware story app for two sisters

Live in production. Push notifications, sibling-aware narration, GDPR-compliant parent dashboard. Started as a problem in my own house — became proof that LLM products can be tender, not tacky.

Next.js · PWA · GPT-4o · Postgres
RESULTS
Livein production
0PII to LLMs
Nightlyhousehold QA
Read full case study →
Tonight's storyThe Lighthouse and the Two Sisters Who Couldn't Sleep
CHAPTER 1 OF 3 · 6 MIN READ
See all case studies → includes shipped & shelved work
How I help

Three ways in. None of them is “hire an agency.”

MODE 01

Advisory call

Single 60-minute working session. You bring the bottleneck, I bring a shipped reference and an honest answer about whether to build, buy, or skip it.

Free · async pre-read · 60 min
MODE 02 · MOST POPULAR

Fractional sprint

2–6 weeks embedded with your team. I scope the system, write the spec, ship the first working version with your engineers, and hand over docs they can actually maintain.

€€ Weekly retainer · capped scope
MODE 03

Build with me

When the system has to actually exist, I scope it, build the first working version, and deliver it production-ready. Milestone-billed, fixed scope, real warranty on the work.

€€€ Fixed-scope SOW
Work with me

If any of this looks like a system you need built or fixed, send me a message.

Message on WhatsApp → Message on Telegram → Reply within 24h, weekdays.
Common questions

Before you write — likely answers to what you're about to ask.

What does an engagement cost?

Advisory calls are free. Fractional sprints and builds are scoped per engagement — the call is where we decide what shape fits, and what it costs.

Do you take equity instead of cash?

Rarely. I'm an operator, not a co-founder. Equity makes sense when an advisory relationship is long-running and the company is past product-market fit. Otherwise, cash.

Where are you based, and does it matter?

Croatia. Engagements are remote-first. I work across European and US business hours; on-site is possible for fractional sprints with a clear scope.

What languages do you work in?

English and Croatian for client-facing work. Code stack is whatever the engagement requires — Python, TypeScript, Kotlin, Go.

How fast do you reply to a first message?

Within 24 hours, weekdays. Weekends I read but rarely reply.

What kind of work isn't a fit?

Anything where the question is “can you make my dashboard prettier” or “can you wire ChatGPT into our existing tool.” I do systems, not surface.

Tell me what's broken. I'll tell you if I can help.

Reply within 24h, weekdays. Croatia time, mostly mornings.
WhatsApp → Telegram → Email →