AI R&D Internship

My mission

At the end of my first year at Télécom, I did a (paid) summer internship at Mellisim.

Mellisim is a french startup funded by the company ValoTec, that aims to create :

• a fully disposable glucose biosensor specifically designed to be affordable for Type 2 Diabetic patients

• a dedicated application on the patient’s smartphone with embedded AI diabetes specialist expertise and coaching skills  to support T2D patients’ specific needs.

When I was in Mellisim, much of the team focus was still on the development of the biosensor and it's "applicator". I was therefore responsible for the R&D work investigating how to develop the dedicated application. I worked in full autonomy, on-site, with weekly reports to my supervisor.

My work

I've signed a non-disclosure agreement with Mellisim for this internship, also I tell here the exact results of my work.

I can frame my role at Mellisim as "investigating algorithms that would be used to give feedback and advise diabetic patients on their (useful and harmful) lifestyle habits".

I divided my work as follows:

• Data cleaning of a clinical trial that Mellisim conducted with diabetic patients: modeling assumptions, outlier detection, missing value imputation, ... Working with python (Pandas). I used this data throughout my whole internship to train and test my algorithms. (1 week)

• R&D of AI algorithms to predict useful insights from patients glucose and actimetric data: when they have meals, snacks or drinks, when they exercise, when they sleep, ... (2 weeks)

• Reading of papers and scientific sources about diabetics to link data insights with useful advice for patients. Implementation of some common metrics used by physicians to assess diabetes evolution. (<1 week)

• R&D of an AI algorithm to detect habits (= patterns) in their eating/exercising routine, and explaining why it recommends that. (2 weeks)

• The last few days: organizing the structure of the documentation I had written during all those weeks, explaining everything I had done/decided/hypothesized, assessing that everything was reproducible, etc. (<1 week)

What I take away from this internship

My supervisor and the CEO were very satisfied with my work. For me, this was my first real professional experience, and the main thing I take away is that my experience working on (numerous) personal projects transfers well to the professional background.

On a technical viewpoint, this internship helped me gain better comprehension of a lot of ML frameworks, of project management, data quality, etc. I considered this internship and the autonomy I had an opportunity for taking the time to understand every little think that I felt I misunderstood.

Gallery

Here are some illustrations of my work at Mellisim: