Jan Sans
Domingo
I turn signal into decisions — voice into filled forms, video into feedback, questions into answers. Computer engineering student building AI pipelines, computer vision systems and full-stack tools that ship.
I'm a Computer Science and Engineering student at the Universitat Autònoma de Barcelona, currently on academic exchange at UC Santa Cruz. My work sits at the intersection of applied AI and systems that have to actually work in the field — from an incident-reporting tool used by emergency departments, to a robot that solves physical puzzles, to a computer-vision system that scores exercise form from video.
I like problems where the hard part isn't the model — it's the pipeline around it: getting messy real-world input into a shape a system can reason about, and turning its output into something a person can actually use.
Outside of coursework, I'm part of the founding team behind FireForm, a UN-recognized Digital Public Good now developed through Google Summer of Code.
FireForm
Emergency responders re-type the same incident across a dozen different agency forms. FireForm turns one voice memo into every required document — automatically.
- 1st place, Reboot the Earth Hackathon (United Nations + UCSC)
- Officially recognized as a UN Digital Public Good; now active in Google Summer of Code
- Led pipeline design & tech stack selection; built the frontend and backend API/DB layer
UCSC Police Department Dashboard + AI Chatbot
Public crime-data dashboard for the UCSC Police Department, extended with a RAG-based chatbot that answers questions in natural language and can filter the dashboard directly from the conversation.
- Fixed and extended an existing Power BI dashboard — interactive map, live filters
- Built the retrieval pipeline: vector search over normalized crime data, answers generated with Gemini
- Data pipeline: RIMS → sanitization → Node/Express on Linode → PostgreSQL → Power BI
Body Posture Analysis System
A computer-vision system that scores squat and push-up technique from 2D video — and the story of replacing a fixed rulebook with a model that actually learns what "correct" looks like.
- Dropped a custom pose-estimation model for MediaPipe once training cost proved impractical
- Rule-based detector alone scored 0% correct reps on real validation video — thresholds were too strict, not the pipeline
- Trained a multi-label Random Forest on the same biomechanical metrics, validated with Leave-One-Video-Out
- Result: accuracy improved to 81–87% per error type; replicated end-to-end for a second exercise
PuzzleBot — Autonomous Puzzle-Solving Robot
A CNC-style robot that sees a jumbled puzzle, works out the solution, and physically assembles it with a suction-cup arm.
- Owned the full software stack: computer-vision piece detection and the solving algorithm
- Generated the robot's pick-and-place command sequence
- Contributed to the physical build and structural design
Managed and maintained kuvings.es, the Spanish e-commerce site for the international Kuvings brand. Shipped updates for performance, functionality and SEO.
Guided students through exam preparation, problem-solving and study planning across school and private tutoring settings.
Let's talk
Open to internships and graduate roles in software engineering, applied AI and computer vision.