Co-founder & cto at Retras
From January 2023 to October 2024 (20+ months)
From January 2023 to October 2024 (20+ months)
Retras aims to motivate patients to go through their handwriting rehabilitation, to solve the current situation where 50% of patients drop out from their rehab.
For that, Retras uses an AI-powered medical device that:
Gives real-time feedback on the rehab exercises
Uses motivating exercises built for adults (and not adapted from children notebooks)
Improves how therapists can monitor their patients' progress and struggles, through analytics and metrics collected at the patient's home, during self-rehab.
The Retras medical device consists of a tablet and a camera, that records the patient's rehab.
I co-founded the project Retras in January 2023. We won the Pepite pitch challenge in February 2023.
After my end-of-study internship, we immediately entered the INRIA Startup Studio in Paris. We received one year of financial support, entrepreunarial design help, HR support, etc. from INRIA.
My role was founder and "CTO":
I helped with designing the company, the business model, the Value Proposition, strategic positioning, etc.
I took part in most presentations and pitchs with prospects, interested investors, therapists, patients. We presented our solution in front of many hospitals in Paris, Lyon and more.
Specifically, as the "technical" guy among the 2 founders:
I was in charge of defining the technical strategy & roadmap
I conceived the whole Retras software and hardware architecture
I managed 7 internships and student projects (ranging from 4 to 5 months)
The first months, our main goal was to validate the Product-Market-Fit and the overall concept of Retras.
Technically speaking, I produced 5 different demos to show clients and partners what Retras final solution might look like: handwriting automated analysis to compute some metrics, a motivating Pictionnary-like exercise with AI audio feedback on the drawing, "sonification" of the handwriting troubles, ...
In parallel, we were willing to explore technical aspects to validate the technical feasibility of the project. I made P.O.C. AI models for different key tasks in Retras medical device:
A letter segmentation model, to cut the handwriting data into letters, and unlock a lot of interesting metrics to report to the therapist
A high-precision pose estimation model to analyse the patients' posture while writing: key to a good rehabilitation according to experts we met
To explore these topics further, I supervised four 4-months student projects:
1 student worked on improving the segmentation model. We achieved +10% accuracy.
3 students worked on the pose estimation model. I split the work between pre-, post-processing and model training. We achieved +5% performance, better temporal consistency, and better depth estimation.
At this point, around March 2024, we were confident on the technical feasibility, and we had good feedback from therapists and from a patient survey.
I kept updating during the whole year a description of Retras technical implementation. My aim was to compromise between rapid prototyping and testing - very important for a startup - and avoiding accumulating technical debt - to be able to quickly ship our MVP (Minimum Viable Product) when hospitals would want it.
We began an Agile development cycle:
Every 2 to 3 weeks, we had some presentation (entrepreneurial events like Spring Saclay, Vivatech, etc., presentations at major hospitals in Paris, Lyon, etc., pre-clinical tests, feedback from therapist partners, ...)
This presentation milestones required stable and well-tested versions of the software and hardware
We developed the features and Proof-Of-Concept functionalities incrementally.
To help us develop the MVP, we recruited 3 interns. I wrote the offers, and carried along all the interviewing process.
Developing the User Interface on an e-ink tablet, with real-time connections with the python backend on the computing device
Developing the audio environment for motivating rehabilitation
Investigating a solution without the tablet, but retaining the same quality of analytics, in case we would need to cut costs.
In the summer 2024, we convinced BPI (Banque Publique d'Investissement) and managed to get a grant to develop the project.
We also developed a version of our solution for data collection as part of pre-clinical tests. I took care to comply with the GDPR (pseudonymization of faces in collected videos by blurring them, patients ID, ...), and I anticipated the technical implementation to one day comply with the european regulation and recommendations on software medical devices.
Finally, I redacted 4 invention reports of the key aspects of our solution.