Skip to work

Davit Simonyan

AI automations

I learned AIthe only way that counts.

Three products live, built at Arag Agency. Inbox Autopilot turns email into actions. Tyoma reads a website, then talks. Where2Pee finds the nearest toilet. I am looking for a team that ships.

products
3live
EN · NL · RU · HY
4langs
available
1builder

Selected work

Not a tutorial dump. Production URLs.

01

Inbox Autopilot

AI that turns email into actions.

The inbox is a pile of unstructured work. This system reads each message and decides what to do — extract, draft, or escalate — instead of leaving a human to sort the queue.

  • Invoice arrives → structured data extracted
  • Inquiry arrives → a reply is drafted
  • Ambiguous mail → routed to a human with a reason
Open liveLive · 2026

Live demo

Click a message

  1. Reading
  2. Classifying
  3. Acting

Working…

02

Tyoma

Your website, already understood.

Paste a URL. In a few seconds you are chatting with an assistant that already read the site. No onboarding dump, no “here is our FAQ.” The context is the page itself. A live product from Arag Agency.

  • Instant URL scan, then grounded chat
  • Interface in English, Dutch, and Russian
  • Built in public as a live demo for Arag Agency
Open liveLive · 2026

Live demo

Scan, then ask

  1. 01Fetch /
  2. 02Read hero + case studies
  3. 03Index 3 live products
  4. 04Ready

03

Where2Pee

Nearest toilet. Still actually there.

You need a toilet now, not a directory. The app locates nearby ones, shows open or closed, free or paid — and lets the last person who used it stamp the listing. Reports land on the record: closed, missing, paid, wrong.

  • Nearby first — distance, hours, price, on the listing
  • “I used this toilet” updates last verified
  • Something changed? Closed, missing, paid, or wrong — the listing takes it
Open liveLive · 2026

Live demo

Schagen · 3 nearby

WC · On the street

Markt, Schagen

Distance
90 m
Price
Free
Hours
Open now (listed hours)
Last verified
2d ago

About

There wasn’t really a master plan.

I’m Davit, 26. I moved to the Netherlands in 2022 and somehow went from working at brands like Scotch & Soda, H&M and Miniso to building AI products in my free time.

I had ideas, wanted to see if I could build them, and got slightly obsessed with figuring it out.

Now some of those ideas are actually live — and I’m curious where I can take this next.

Based
Schagen · Netherlands since 2022
Languages
English · Dutch · Russian · Armenian
Academy
Build with Claude — Huva.io Academy (completed)
Also
Scotch & Soda · H&M · Miniso flagship
Stack
Claude API · n8n · Supabase · Groq · DigitalOcean · Netlify · Vercel · Claude Code · ChatGPT · Codex · Grok

Process

How the work gets made.

    01

    Tight loops

    Models, prompting, tool use, evaluation. Fewer notebooks. More loops: try, measure, throw away, try again.

    02

    Ship the agent

    Inbox Autopilot. Classification, extraction, a human in the loop. The point was not a demo — it was a system that acts.

    03

    Ground the chat

    Tyoma. Fetch a site, ground an assistant, converse. Three languages. A public URL, because private slides do not hire you.

    04

    Make it speak

    Polish, evaluation, this page. If the work cannot introduce itself, a resume will not do it either.

What I actually do

Agentic workflows

Read → decide → act, with a human gate when the model should not be trusted.

Extraction

Unstructured mail and documents into fields you can actually book, file, or pay.

Site-grounded chat

Assistants that start from the page, not from a hallucinated brochure.

Product, not a gist

Live URLs. Loading states. Languages. The unglamorous parts that have to work on Monday.

Recruiters · hiring teams

Davit Simonyan · AI automations · Arag Agency · Schagen, Netherlands

Hiring? Let’s talk.

Want to discuss a role or my experience? Pick a time that works for you.

Write

Not a recruiter? Write here — it lands in my inbox.