I build production AI systems people trust and love.

I'm Mykel Schmidt, a hands-on Lead AI Engineer at Flynth. I build secure AI applications on Azure, from the architecture down to the code, and I design them around the people who will use them.

Download résumé LinkedIn

Portrait of Mykel Schmidt — **ir. Mykel Schmidt** MSc Human-Technology Interaction

Lead AI Engineer at Flynth

I joined Flynth, a Dutch accountancy and advisory firm, in November 2025 to help build its AI platform. Today I lead the platform and the solutions built on it.

Nov 2025 AI Engineer

May 2026 Lead AI Engineer

Python
FastAPI
PostgreSQL
Docker
Azure
Microsoft Foundry

KeyTakes

My own AI product, since June 2025

KeyTakes summarises any webpage or YouTube video with Google's Gemini models. You can also ask it follow-up questions. A RAG pipeline looks up the relevant parts of the source and hands them to the model.

I built all of it myself. Async serverless functions on Vercel stream the results to a Next.js frontend over Server-Sent Events, Puppeteer does the scraping and Firebase handles accounts and state.

Next.js
Gemini
RAG
SSE
Puppeteer
Firebase

Visit keytakes.com Summarise this site with KeyTakes

Data & AI Engineer at E-mergo

A data and analytics consultancy, May 2024 to May 2025

Flex IT Rent Multi-agent RAG

I led the discovery, the architecture and the proof of concept, built with Python, Azure AI Foundry Prompt Flow, AutoGen and web scraping.

Client evaluation Maximum score

The client gave me the maximum score in their evaluation and approved the next phase.

Eindhoven Airport Data warehouse

I helped set up their TimeXtender data warehouse: databases, API connections and SQL transformations for high-volume streaming data, incremental loads and changing dimensions.

I also built reusable patterns for API ingestion and incremental loading, and migrated Qlik transformations to TimeXtender.

Dawn

New tab dashboard for Chrome, since 2023

After my Master's I took a sabbatical and built Dawn on my own. It replaces Chrome's new tab page with a dashboard for focus and wellbeing. You choose the cards and put them where you like, and an LLM powers one of the features.

10,000+ downloads on the Chrome Web Store

Featured by Google

Chrome Web Store Try it on the web

Dawn with its default cards: positive thinking, a daily mantra and a morning routine

Dawn with habit cards: day counters, long-term goals, a vision board and the X-effect

One of the 40 trials from my experiment

Same person, or two different people?

First photo: a man with a shaved head in a grey shirt

13

Second photo: a man with long hair, a beard and tinted glasses

My experiment: the AI's confidence sets how long you get to decide

Three of the four groups got a fixed 13 seconds per trial. In the fourth, the AI's confidence set the time:

95% sure 9 s
75% sure 13 s
55% sure 17 s

t = 8 + \frac{100 - c}{5}

where $t$ is the time to decide in seconds and $c$ is the AI's confidence in %

What if the AI's confidence set your time to decide?

My Master's thesis at TU/e, 2023

In a deferral system the AI makes a first call and a person makes the final one. I proposed letting the model's confidence decide how much time that person gets: less when the AI is sure, more when it isn't.

I built the experiment as a web app with 40 timed trials and ran it with 266 participants in four conditions. Showing a confidence percentage made people slower without making them more accurate. My own framework didn't give a significant gain either, and I reported that null result as it was.

Read the thesis

Eindhoven University of Technology

Both my degrees are about how people and technology work together.

MSc Human-Technology Interaction

8 or higher in 80% of my courses. My thesis was about how people and AI make decisions together.

UX Design
Advanced Cognitive Engineering
Human-Robot Interaction
Management of Product Development

BSc Psychology & Technology, ICT track

I picked the ICT track because I love gaming and computers. It gave me programming, databases and research methods next to the psychology.

Programming
Data modelling and databases
Logic and Set Theory
UX for Intelligent Systems

What I'm good at

AI product engineering

Production LLM apps with multi-stage and multi-agent workflows, RAG, structured outputs, human feedback, evaluation and control over cost and latency.
Software and cloud architecture

Async Python services and APIs on Azure, with containers, infrastructure as code, CI/CD, identity and security designed in from the start.
Reliable and responsible AI

Source grounding, deterministic controls, automated testing, model benchmarks, privacy by design, DPIA/TIA work and human oversight.
Technical leadership

Architecture and product decisions, engineering standards, hiring, and translating between the business and the tech, while staying hands-on.
Machine learning and data

Classical ML next to LLMs, statistical analysis and experiment design, data warehousing and BI.
Human-centred design

A background in psychology, UX and research that I use to make complex systems understandable and useful.