起動 · Boot sequence
Sound on ♪
غازي شريف
Tap the ground to throw a ball
Stride 0.57 BL · no slip
INU-01 is an animated robot dog. Throw it a ball, make it run to music, or watch it do graded push-ups.
EPFL for the bachelor, a year at Sorbonne in Paris, then the MSc at ETH Zürich in Robotics, Systems & Control. The coursework that stuck: deep learning for autonomous driving, model predictive control, robot dynamics.
Master focus project: I added hand movement data to gaze-based action recognition. Task recognition accuracy rose to 91%. Graded 5.5 out of 6.
Mechatronics intern: automated test frameworks, embedded systems and CAD hardware, plus OCR computer vision that had to work in real conditions, not in a lab. On the side, an Arduino RC kart.
Freelance lead mechanical engineer: I designed automated production machinery and worked with manufacturing partners until it ran in real-world operation.
Master thesis: an RGB-D method to guide pedicle screw placement in robotic spine surgery. ZED stereo cameras, optical tracking, segmentation of the lumbar spine from L1 to L5, and live tracking in ROS2 for the EU Horizon FAROS project.
Idea to shop floor with full P&L: budget, permits, suppliers, equipment, staff and SOPs. Up to 200+ units a day for 7 B2B clients and walk-in customers, at about 65% gross margin.
Founder Associate at an AI startup that audits freight invoices against negotiated tariffs. The full enterprise sales cycle across Switzerland, an automated go-to-market pipeline that booked inbound meetings, and 40 to 80 cold calls a day.
The gym, measured. I co-founded the company behind two live AI fitness apps: HyperBody, a physique coach that scores every muscle group and rewrites your plan every week, and Rate My Body, an aesthetics score from 0 to 100. HyperBody's form checks run on the analysis engine I wrote.
Season finale. Commercial operations and client enablement at a life sciences quality consultancy in Basel, reporting to the CEO. Nine projects so far, from an engine that reads FDA warning letters every night to an AI that drafts validation documents a qualified person signs.
Measured an observed result Designed how it is built, not a promise
次回予告
Next time on INU-01
Starring INU-01, Ghazi Cherif and you
You found it. Nobody ends up here by accident.
Everything is synthesised live. Nothing here is a recording.
Found ones show what they are. The rest only give a hint.
No framework, no build step. One HTML file for the portfolio and one script for the dojo, loaded only when you open it.
INU-01 is redrawn on a 2D canvas every frame. Each leg is a two-link chain, and inverse kinematics finds the knee from the hip and the foot target. Gaits are phase offsets per leg, the same model used to describe real legged robots. The footfall plot shows which feet are on the ground.
The music is not a recording. A Web Audio synthesiser plays the royal road by day, city pop on the 丸サ進行 at dusk and lo-fi at night, scheduled ahead on the audio clock so the feet land on the beat. Tap the clock up top to change the sky: the workout, the toy and the music change with it.
Every project page has a tiny demo. Illustrative toys with sample data, drawn on canvas and loaded only when you open a project, so the page itself stays light.
The dojo's rep counter runs Google's open-source MediaPipe pose model in your browser. Same 33 landmarks as my form analysis engine. Your camera feed never leaves your device.