An NFC and QR card that sits on a business's counter. A customer taps it with their phone and lands on the business's Google review page.

What it took: I built the whole product in about a week: the card design, the website, logging every tap, and a dashboard where the owner sees taps and reviews. Unhappy customers can leave private feedback, so the owner hears about problems and can fix them. Emails go out automatically through Resend, and the data lives in Supabase.

How I sold it: I made a list of 373 Latvian businesses and scored them by priority, then messaged businesses I didn't know on WhatsApp. The first 10 cards went mostly to clinics, plus tow and car transport companies, and I invoiced them myself.

Open reviewtap.me

What it took: I set up my own Google Workspace sending inboxes on my own domains, with SPF, DKIM and DMARC, and warmed them up for three months before sending.

I find and qualify leads with data from Treg (a prepaid data API) and Apify scrapers, send through Salesforge (before that Instantly), and track every lead and reply in Twenty CRM, with a booking link for calls.

Now building: an AI SDR, a team of agents that finds and checks leads, writes each message, handles replies and books calls.

What it took: I built generators that fill in the firm's client documents as Word forms, and Inbox Clerk, a tool that reads an emailed order or quote, turns it into rows for the accounting system and drafts the reply.

Government register checks stay with a person on purpose, because the rules expect a human to do them.

Result: about 15 hours of work saved every month.

What it took: a scanner that reads Latvia's open government data (tax records, the business register and filed annual accounts) and estimates where each company is losing money.

Result: 3,779 companies with 5 to 80 staff mapped at €0 data cost, so a first message can start from a number about their own business.

What it took: a lead agent takes a goal and splits it between scouts that research, builders that produce and a council of agents working in parallel. An independent reviewer agent checks the result before I use it.

The agents use MCP servers to reach real tools and data, and a knowledge base I made from about 90 business books.

So far: 7 custom agents, 88 Python tools, and 2 add-ons for Claude Code itself: a live panel that shows what each agent is doing, and a meter that shows what a session costs.

After a meeting with an executive search firm, I prepared a one-page proposal in Latvian with three AI solutions for their team.

In 11th grade I learned investing, saving and entrepreneurship, gave 8 lectures on money to other students and finished at the top level, Everest. The same year I shadowed the programme's director. This school year I'm back as a mentor for new ambassadors.

Read my article on seb.lv

Took part in the youth innovation forum at Vidzeme University of Applied Sciences on 27 February 2025.

AI: Claude Code, MCP servers, sub-agents and agent teams, evals and verification loops, skills, computer and browser use, scheduled tasks, RAG and vector search, knowledge bases for agents, context and prompt engineering.

Building: APIs and webhooks, Python, Git and GitHub Actions, VS Code, Supabase, Resend, Zapier.

Sales and data: Google Workspace admin, inbox warm-up, SPF, DKIM and DMARC, Salesforge, Instantly, Twenty CRM, Treg, Apify.

I taught myself to train: five years of weights, three years of running, a Rimi Riga Marathon under four hours and three years of volleyball. I built my own home gym and taught my cousin to train. Sport taught me that results come from doing the same thing every day, even when I don't feel like it.

I've read 20+ books on psychology, sales, self-improvement and finance. The two that shaped me most: Can't Hurt Me and The Success Principles.