Skip to content

AI Engineer · Product Owner

Bangkok · Remote · EU citizen · Open to full-time roles

Raised in acar workshop.Now I buildAI agents.

I'm Adrian Szramowski. I started on the workshop floor, then served the customers, then ran the team. Today I build AI agents that run in production. One I built and maintained handled 1.5K conversations in 30 days.

Adrian Szramowski in a black shirt, hands on his hips

Read off the dashboards

  • 1.5Kconversations in 30 days on one inbound voice agent, 99.9% success rate
  • 82/144crew check-ins logged by an AI agent in one week, the share the client asked for
  • 990live calls in one review view, each with outcome, sentiment and a review flag
  • 150+forex brokers on the crypto payment platform I owned as Product Owner

Screenshots below. Client names are hidden.

01How I work

I start with the business.

Plenty of people can wire up an AI agent. Fewer ask what it has to earn before they build it. That habit comes from years behind a workshop counter, where every job had a customer waiting and a price on it.

  1. 01

    Ask what it's worth

    Before I build, I find out what the problem costs and what fixing it pays. A US solar contractor wanted a set share of crew check-ins off their operator. I built the agent, then a weekly report that shows the share.

  2. 02

    Know when to hand over

    An agent shouldn't try to handle everything. On the solar check-in line, urgent calls still go straight to a live operator. On the US voice agents, a caller who needs a person gets a ticket with their reason and a summary, so staff pick it up.

  3. 03

    Test it like the customer will

    An agent that works in a demo can still fail on a real call. I reviewed live calls every day: outcome, sentiment and a review flag on each one, 990 calls in one view.

Call review board: date, duration, outcome, sentiment score, review flag and ticket for each call, 990 matches
The call review board I worked from every day. Caller details blurred.

02Selected work

Built, shipped, measured.

  1. 01 · Voice agents · US businesses · 2026

    Inbound voice agents that answer, book and hand over

    Problem
    Small US businesses were missing calls or logging them by hand: a trades company, an auto dealer, a chiropractic clinic.
    What I did
    I built and maintained an inbound agent for each, with its own prompt, knowledge base and booking or ticket flow. I reviewed calls daily and got SMS through US carrier registration (A2P 10DLC).
    Result
    • 1.5K conversations in 30 days on one agent, about 12 cents each
    • 277 calls in 30 days for the auto dealer, 97.1% connected
    • 124 calls in 30 days for the clinic, 98.4% connected

    ElevenLabs · Vapi · Twilio · GoHighLevel · n8n

    ElevenLabs dashboard: 1.5K conversations in 30 days, average duration 1:14, average cost $0.123, success rate 99.9%
    One agent, 30 days, ElevenLabs dashboard
  2. 02 · Voice and SMS agent · US solar contractor · 2026

    Crew check-ins without an operator typing them

    Problem
    Field crews phoned in to check in and out of jobs. A live operator took about 500 calls a month and typed each one into Salesforce.
    What I did
    I built a voice and SMS agent that identifies the crew and the job and writes the record to Salesforce through n8n, with reminders for crews who haven't checked in. Urgent calls still go to a person. A weekly report shows the client agent work against manual work.
    Result
    • 82 of 144 check-in logs created by the agent in one week (56.9%)
    • 28 by voice, 54 by SMS

    Vapi · Twilio · n8n · Salesforce

    Weekly report: 82 check-in logs by the AI agent, 62 typed by a person, 144 in total, agent share 56.9%
    The weekly report the client gets
  3. 03 · Product Owner · Match2Pay · 2023 to 2024

    A crypto payment platform for 150+ forex brokers

    Problem
    Brokers need crypto deposits and withdrawals that just work. Every gap shows up as a failed deposit and an unhappy trader.
    What I did
    I owned the product: gathered requests from business clients, wrote the specs the developers built from, planned the roadmap and directed a full frontend and backend rebuild of the platform over one year.
    Result
    • 150+ forex brokers on the platform
    • Full rebuild, frontend and backend, in one year

    Jira · specs · roadmap

    Match2Pay crypto transactions dashboard: deposits and withdrawals per month and by coin, amounts hidden
    Match2Pay dashboard, amounts hidden
  4. 04 · Project · AI visibility

    Meristoll: does AI recommend your business?

    Problem
    More buyers ask ChatGPT or Gemini who to hire. Most small businesses have no idea whether they get named.
    What I did
    I built a report engine with Claude Code (Next.js, Supabase). It asks 6 AI assistants the questions a business's buyers ask, then turns the answers into a report with a grade and the competitors AI names instead.
    Result
    • 6 assistants: ChatGPT, Claude, Gemini, Perplexity, Grok, Google AI Mode

    Claude Code · Next.js · Supabase

    meristoll.com
    The meristoll.com homepage
    meristoll.com

03Tools I built for my own work

When a task repeats, I build the tool.

  • Lead scoring

    Reads each business website with an LLM, checks every fact against the page and scores the lead 0 to 100 against a customer profile.

  • Brief to demo

    Turns a client brief into a working demo, a client presentation and three proposal drafts, with a quality check before anything goes out.

  • Call review dashboard

    Clients sign in and see every call their agent handled: transcript, recording, caller and the outcome the agent pulled out.

  • Reel editor

    Turns a raw talking-head recording into a finished Instagram reel with cuts, captions and B-roll, with four approval gates on the way.

05Experience

The path so far.

  1. 2026 to now

    AI Engineer, independent

    as Fast Growth AI · remote

    Voice agents, SMS agents and automations for US businesses, directly and as a delivery partner for a US AI agency. I scope the work with the client, build it, test it and report the result back.

  2. Oct 2024 to Dec 2025

    Dealing and Broker Support Specialist

    Match-Trade Technologies

    Main technical contact for forex brokers on the trading platform. Helped partners integrate the platform API and kept their service tickets moving.

  3. Jun 2023 to Oct 2024

    Product Owner

    Match2Pay

    Owned a crypto payment processor used by 150+ forex brokers. Requirements from business clients, specs for the dev team, the roadmap and a full platform rebuild.

  4. Jun 2022 to Jun 2023

    Junior Product Support Specialist

    Match-Trade Technologies

    Supported brokers and their traders on the trading platform and the CRM.

  5. Before 2022

    Mechanic, customer service, team lead

    Mechanika Szramowski, family workshop · Elbląg

    Holidays and most weekends for as long as I can remember. Fixed cars first, then took care of the customers, then ran the team.

  6. 2018 to 2021

    BA in Finance and Accounting

    WSB Merito University, Gdańsk

Polish (native) · English (fluent, daily calls with US clients)

Contact

Hiring for AI or product?

Send me the role and a line about the team. I'm open to full-time work, remote or with relocation.