Oct 2026
Running my models in the browser
How a scikit-learn MLP and a YOLOv8 segmentation model ended up running on this site with no server, and what it took to make their outputs match the originals.
Duarte Pereira da Cruz
I build machine learning systems, mostly in computer vision, and I care as much about running them well as about training them.
Now
I’m in Copenhagen doing an MSc in Autonomous Systems at DTU, and from November I’m also working part-time as a junior AI consultant at Implement Consulting Group. Before that I studied Computer Science at Técnico in Lisbon and spent a summer on Cloudflare’s developer infrastructure team.
I’m looking for an MSc thesis topic or a research collaboration in perception and machine learning. If that’s something you work on, get in touch.
Selected work
- 2025–26Detecting Portuguese fake newsFive classifiers trained on 63,000 European Portuguese news articles, and a close look at why they decide what they decide.
- 2025Mapping invasive acacias from aboveA YOLOv8 segmentation model that outlines invasive acacia in satellite imagery, trained on a dataset we built from Portugal’s land-use map.
- 2025A meeting assistant for AR glassesLive captions, a running summary and automatically extracted tasks, in your field of view on Snapchat Spectacles.
- 2026Unscrolling, a news podcast made for one listenerAn agent that reads the sources you follow and turns everything new into a few minutes of audio, so you can stop doomscrolling for it.
Earlier
- EY AI Challenge2025A Neo4j graph of how Portugal’s Labour and Civil Codes reference each other. Selected from more than 500 applicants.
- CanSat Portugal2023Led the team building a satellite the size of a soda can, with a cork shell and a thermoelectric generator that turns the electronics’ waste heat into power. National finalist, top 10.
- Astro Pi2023Led four developers running a 24-hour simulated ISS isolation study on Raspberry Pi. Exhibited at Ciência Viva.
- Sports data appA Flutter app on Firebase and the SportMonks API.
Writing
Experience
From Nov 2026
Implement Consulting Group
Junior AI consultant, The Tech Collective (Data & Analytics), part-time
- Joining Implement’s data and analytics unit to work on client projects alongside my MSc at DTU.
Jun – Aug 2026
Cloudflare
Software engineering intern, Developer Infrastructure
- Main project: the Fuzz Testing Platform, which runs long-running fuzzing jobs on Cloudflare’s internal container infrastructure. Co-designed its off-cluster gRPC control plane, and wrote the client and two-way Kubernetes reconciliation in Go. Also built a TypeScript provisioning shim on Durable Objects and a reusable GitLab CI component other teams use for fuzz campaigns. Over forty merge requests merged in three months.
- On my own initiative, tracked down a concurrency race in the shared OAuth token-refresh path of an AI-agent gateway used internally and by customers. The fix, a SQLite-backed single-flight leader election, removed about 9,700 errors a week, 97% of the gateway’s failures.
- Hardened platform APIs against race conditions and cross-tenant leaks, backed by more than 490 unit tests.
“He did a lot of real work in a short time (over forty merged merge requests) and the quality was consistently high. Duarte is self-directed, learns fast, and looks for problems worth fixing rather than waiting to be told.”
Jan – Oct 2026
JunctionX Lisbon 2026
Co-lead organiser, with Daniel Borges
- A year after winning JunctionX Lisbon, I co-led the next edition: 150+ participants at Técnico’s Taguspark campus on 10–11 October 2026, with 24 hours of hacking and more than €25,000 in prizes.
- Ran an 8-person organising team under the student group Gradient, and pitched the event to sponsors and API partners. Four companies set the challenges (NOS, Sensei, Inspiring Solutions and Albatroz Engenharia), with support from Jane Street, ElevenLabs, Lovable, n8n and MongoDB, among others.
Feb – Apr 2026
Instituto Superior Técnico
Teaching assistant
- Ran lab sessions for the Human-Computer Interaction course, a class of 24.
Jul – Aug 2025
Nmbrs, Visma Group
Tech intern
- Turned the team’s architecture conventions into Markdown and YAML rule sets that let LLMs generate complete Data, Core, IoC and API projects.
- Automated project scaffolding, cutting setup time by 87%.
“He created smart boilerplate generation that reduced our typical 5-day setup process to about 1 hour. What made this particularly impressive was that Duarte wasn’t familiar with our internal guidelines and wasn’t an advanced .NET developer when he started.”
Education
2026 – 2028
Technical University of Denmark
MSc Autonomous Systems
Perception for Autonomous Systems, Deep Learning, Deep Learning in Computer Vision, Robotics, Machine Learning Operations.
2023 – 2026
Instituto Superior Técnico, Lisbon
BSc Computer Science and Engineering
GPA 16/20. Academic Merit award, 2024/25.
- Machine Learning
- 20/20
- Human-Computer Interaction
- 19/20
- Probability and Statistics
- 18/20
- Theory of Computation
- 18/20
- Artificial Intelligence
- 17/20
- Analysis and Synthesis of Algorithms
- 17/20
Tools
- Machine learning
- PyTorch, scikit-learn, YOLOv8, Hugging Face, LIME, Gemini and ElevenLabs APIs
- Languages
- Python, Go, Rust, C and C++, TypeScript, Java, C#, SQL
- Infrastructure
- Kubernetes, gRPC, Docker, CI/CD, FastAPI, Celery, Redis, Postgres, Neo4j, AWS S3
- Spoken
- Portuguese (native), English (C1, IELTS 8.0)