I build AI products and explain how modern systems actually work.

Portrait of Ahmed Mahfouz — Software developer at Careweb since 2024. Before that, robot vision at VOIAR.

YearWorkField

2024–Careweb assistant: speech to draft care notesHealthcare AI 2023–24Vision for a person-following forest robotRobotics 2024TensorRT image for YOLO on JetsonOpen source 2026Qwen3-ASR for WhisperXOpen source 2024–ListiMate, Storify, Couples QuestionsOwn apps 2019–23N.E.A.T game AI, Cyclone arcade gameEarlier

contact@ahmedmahfouz.me Download CV

2024 – now · Careweb, Utrecht

An AI assistant inside mental-healthcare software

Clinicians record a conversation. My Python services turn it into a draft care note and appointment suggestions, and the clinician reviews every draft before anything is added to the main Java application.

Role: Software developer
Stack: FastAPI, Whisper large-v3-turbo, WhisperX, pyannote, vLLM, React

Read the case study

1Recordinga conversation in the clinic
2TranscriptionWhisper, WhisperX
3Who said whatpyannote
4Draft note and appointmentsLLM served by vLLM
5Clinician reviewnothing is added without it

A 49-minute recording, transcribed (one local measurement)≈ 160 s

2023 – 2024 · VOIAR

Vision for a forest robot that follows a person

A robot that follows a person through the forest. I built its vision on an NVIDIA Jetson: detection and tracking that keep working on uneven ground and in changing light, tested in the field.

Role: AI developer
Stack: YOLOv8, TensorRT, Docker, NVIDIA Jetson Nano

Read the case study

The forest robot during a field test, with a laptop running the vision code — Fig. 1 Field test, with the vision running on the laptop beside it. Fig. 2 The circuit box.

The robot's circuit box — Fig. 1 Field test, with the vision running on the laptop beside it. Fig. 2 The circuit box.

Open source

Open-source work on Ultralytics YOLO and WhisperX

My TensorRT Docker image for NVIDIA Jetson is part of the official Ultralytics YOLO repository. For WhisperX I have a pull request open that adds Qwen3-ASR and the Qwen3 forced aligner.

Read how the image was built

43%lower inference latency on Jetson, about

72%more frames per second, about

The stock Jetson Nano image, without GPU acceleration, against my GPU-accelerated TensorRT image. Measured for the Ultralytics pull request.

TensorRT Docker image for YOLO on JetsonUltralytics →Qwen3-ASR and forced aligner, pull request openWhisperX →

2024 – now · Own apps

Three apps I designed, built and shipped

From the first sketch to the stores, and every update after.

ListiMateSmart grocery list

StorifyA story from a photo

Couples QuestionsA question game for two

Writing

A TensorRT Docker image for YOLO on JetsonArticle Adding Qwen3-ASR to WhisperXArticle Survive the line: a game AI with N.E.A.TArticle The newsletter on BeehiivNewsletter

About

I'm Ahmed, an applied AI engineer in Utrecht. I like the part of AI that has to survive the real world: privacy rules, a small GPU on a robot, code that thousands of people depend on.

I studied Computer Science in Hungary on a Stipendium Hungaricum scholarship and finished first in my class. I work in English and Dutch teams.

Now: Careweb, 2024 –
Before: VOIAR, 2023 – 2024
Education: BSc Computer Science, 2023