Mees van Berkel

Applied AI and backend engineer.

I build backend services and the systems around AI models. At Hypersolid I built the live transcription pipeline behind Omni, an on-premise meeting assistant that transcribes live audio in about a second.

Availability
Available immediately for software, backend, full-stack and AI roles
Based
Alkmaar, NL, open to remote
Education
BSc Software Engineering, HvA, January 2026
Languages
Dutch (native), English (fluent), Spanish (basic)

Experience

Creative Technologist Intern · Alkmaar, Netherlands

Engineered core components of Omni: live transcription, speaker identification and local-LLM summaries, all on-premise

Software Engineering Intern · New York, USA (Remote)

Survey platform for Columbia University, a Crizit client

Hogeschool van Amsterdam

Graduated Jan 2026

BSc Software Engineering

Thesis graded 9/10

Selected work

Omni

Graduation project, 2025–26

Transcribes live meeting audio in about a second and summarises it with local AI models, so no meeting data leaves the company network.

0.93 s
from audio chunk to text on screen, on average
9.1 %
word error rate on Dutch and mixed Dutch/English meetings

Thesis evaluation: 8 sessions, about 100 minutes of audio, 5 speakers.

My graduation project at Hypersolid, where I owned the transcription pipeline, meeting-data orchestration, search and dashboard.

Read the Omni case study

LinuxWhisper

Personal project, 2026

Offline dictation for Linux: hold a shortcut, speak, and the text is on your clipboard. It works end to end on KDE Plasma.

1.1 s
from letting go of the shortcut to text on the clipboard, for a 1.6-second recording with the base model on CPU
50+
automated Rust and Python tests

Built it end to end: the Rust daemon, the Python speech worker, the JSON contracts between them and the Tauri status window.

Read the LinuxWhisper case study

IoT Power Reel Wall

Team project, 2024

An NFC-locked wall that lends power reels to students. It was used in HvA labs, where every check-out and return goes through the API and shows up in one dashboard.

On a 7-person team, I built the Nuxt check-in and check-out API, using atomic Prisma transactions, and the usage dashboard. I also containerised the services with Docker and worked on the ESP32 NFC firmware.

Built with an ESP32, a PN532 NFC reader and a solenoid lock; Nuxt, Prisma and Docker on a Raspberry Pi.

Read the IoT Power Reel Wall case study

All projects

Writing

Building an honest measurement loop around a game: fair bots, paired seeds, a held-out test, and an AI agent doing the implementation.

Directing an AI coding agent through a large systems port, with the original hardware build as the test oracle.

All writing