Builder · Budapest · since 2011

Náray E. Tamás

Wood, chocolate, grass, concrete.
Code ran through all of them.

Write to me

  1. The next 24.hu apps

    24.hu

    Year
    now
    Status
    Building
    Note
    Research-building the next generation of products, with discovery done in code, not tickets.

  2. orszagjaras.24.hu & tippjatek.24.hu

    24.hu

    Year
    2026
    Status
    Shipped
    Note
    Social game sites for the election and the World Cup. 80–90% of discovery, design and code, solo, for millions of page views.

  3. Környékradar

    Geo data

    Year
    2026
    Note
    Scorecards for all 23 Budapest districts and 80 towns around them, on Postgres/PostGIS and H3 hexes. Built by Haiku/Sonnet swarms, checked by judge agents.

  4. Ügyintézési katalógus

    RAG

    Year
    2026
    Note
    A chat-first guide to Hungarian bureaucracy: 11,000+ offices, 5,700+ procedures and a RAG pipeline honest about what has gone stale. 130+ guides fact-checked by Haiku/Sonnet swarms, with Opus verifying.

  5. Pip

    iOS

    Year
    2026
    Note
    A wordless bedtime game for five-year-olds. One session is one night. No streaks, no currency, no “one more level”.

  6. An agentic editorial tool

    24.hu

    Year
    —
    Status
    Shipped
    Note
    Agent orchestration tuned for mission-critical accuracy and speed. ~60% newsroom adoption in two weeks.

  7. Claude Code, company-wide

    Culture

    Year
    2025
    Note
    A user since release, I made the case until Central Media bought in. Now mentoring teams on it: 50 seats and growing, around 80% weekly active.

  8. Sticker Exchange

    Marketplace

    Year
    2025
    Note
    A swap marketplace for families collecting sticker albums. Built for a household that needed it.

  9. Rebuilding kremmania.hu

    Lesson

    Year
    2025
    Status
    Failed
    Note
    It failed. Agents need every system in the stack, or strictly greenfield. Nothing in between.

  10. A private podcast service

    Software · Hit Radio

    Year
    2021
    Note
    Took a subscription podcast product from business case to launch, stitched together from off-the-shelf, open-source and low-code parts.

  11. A multi-purpose sportshall

    Construction · INTER Academy

    Year
    2020
    Note
    Held the contractors, the private owners and the municipality in the same room, and financed the build. Finished on time and on budget, through a pandemic.

  12. A football academy of 700+ players

    Organisation · INTER Academy

    Year
    2018
    Note
    Rebuilt the organisation on new CRM and ERP systems across three cities. Among the top five INTER Academy projects worldwide.

  13. A chocolate and waffle bar

    Hospitality · Ahoy!

    Year
    2014
    Status
    Sold
    Note
    Built the restaurant and the service around it with service design principles, down to custom software that tracked the guest life cycle.

  14. Wooden phone covers

    Product · Captain’s

    Year
    2013
    Status
    Sold
    Note
    Designed the product and wrote the software that sold it: a full line of covers milled from real wood, with a configurator that kept the choices small enough to decide.

  15. Playgrounds, sold online

    E-commerce · Pull

    Year
    2011
    Note
    Developed the e-commerce portfolio for a home improvement and gardening chain, using conversion rate analysis and content-driven commerce.

The first essays are on the bench.

RSS

How I work

A studio, not a pair.

I run Claude Code the way a technical director runs a studio: whole builds handed to orchestrated agent swarms, roughly 400 sessions across newsroom and personal work.

  • 699commits shipped
  • 399agent sessions
  • 27agents in one swarm
  • ~59%cost cut by tiering

  • Orchestration, not pairing

    Terse, outcome-level asks, then a long run. The swarm does the work; I set the end state.

  • Model-tiering by design

    Cheap models build and fact-check, stronger ones verify. One audit: ~59% cheaper, no quality loss.

  • Evidence over optimism

    A 5-year backtest killed a strategy I liked. Adversarial tests catch real bugs before ship.

  • Ship and verify

    Plan, build, deploy, confirm it is live, log it. Discovery happens in code.

Contact

Got a half-formed idea?

Let’s find out whether it can exist. Open to senior product and technology leadership roles, advisory, and conversations about agentic builds.

Thanks — I read everything myself and will reply within a few days.