Two engineers, one accountable unit.Specialised models instead of LLM wrappers.
AIM: 1 min of 720p analysed in ~7 s on one cloud GPU (was ~6 min)
Hansel: specialised models, metrics not public
Two systems we built that run every day. Every number comes with the conditions it was measured under.
Metrics not publicHansel's numbers stay with Hansel. We show the gap instead of guessing.
Detection, segmentation, pose and tracking that hold up on real video, not only on benchmarks.
Models that run fast on the hardware you already have, on-device and at the edge, with TensorRT, ONNX and CUDA.
Retrieval and agent workflows that stay accurate in production, shipped with evaluations and monitoring.
AWS and Google Cloud with EU data residency, plus the pipelines, deployment and monitoring to run it all.
Specialised models where a general-purpose LLM is the wrong tool, and smaller models that keep their accuracy.
Documentation, logging, human oversight and evaluations built in, so a system can pass a security review or an EU AI Act audit.
Every engagement starts with a small, scoped first phase: your data, your constraints and a measurable definition of done, before any bigger commitment.
The same two engineers take it from raw data to a deployed system: data, models, evaluation and cloud. No handoffs, no rotating bench.
Monitoring, documentation and EU data residency by default, so the system keeps working after we step back.
We reply within 24 hours, in English, Portuguese or German.
AI/ML engineer and strategist
Computer vision, edge deployment and custom deep-learning architectures. Leads the ML pipeline behind AIM. Started in industrial software, writing C and C++ protocol gateways for power grids (IEC 61850, IEC 60870, TASE.2), then moved into ML through bioimage research at the University of Algarve and an MSc in AI at the University of Freiburg, where he worked with the AutoML lab on scaling down TabPFN.
ML engineer and cloud architect
Production ML and cloud engineering. Almost three years at Deloitte building Python backends, data pipelines and DevOps on AWS, with smart-sensor integration, distributed MongoDB and containerised architectures on Docker and Kubernetes. AWS Certified Solutions Architect and Developer (Associate). Now a production and machine-learning engineer alongside Jude. Studied computer science at the University of Algarve, where he founded a programming club and competed at MIUP, Portugal's inter-university programming marathon. Studying for a Master's in AI at the University of Coimbra.
Tell us what you are building and where it is stuck. We reply within 24 hours.
