Amo is a startup that aims at developing the new generation of social media. I was part of the back-end rust team and worked with an architecture designed for millions of requests per second.

At Tarides, a start-up developing open-source softwares, I worked on a DBMS called Irmin which is a critical component of the Tezos blockchain.I was a tech lead in the development team of Irmin. We worked publicly on GitHub and code reviews occupied a significant amount of my time. I also spent a lot of time syncing with my teammates and guiding them.Irmin has a rich feature space and contains a lot of legacy code that often requires urgent improvements to be implemented with a high degree of backwards compatibility. One of my main concerns was to preserve the quality of the code under these challenging circumstances. As an example, I have implemented Merkle Proofs in Irmin: https://github.com/mirage/irmin/blob/main/examples/merkle_proofs.mdEnhancing the performance (i e., runtime, RAM and disk space) of Irmin for Tezos was central to my work. I spent a lot of time writing benchmarks and conducting analyses of Irmin’s performance. As an example, I conducted an important analysis of Tezos disk access patterns that is now used to select the development targets for Irmin: https://github.com/Ngoguey42/segment_hangzhou/blob/master/index.ipynb

Airware was a start-up that made data acquired by surveying drones available via an online platform for clients such as insurances, mines and quarries. I was part of the data science team where my main mission was to research and develop image processing algorithms and deploy them in production.I became proficient with many data-science libraries (e.g., numpy, opencv, keras, pytorch, pandas, networkx), I developped one (see url below), and I trained many CNNs for semantic segmentation (see url below), instance segmentation, image inpainting (with GANs), and multitask learning.buzzard library: https://github.com/earthcube-lab/buzzardsemantic segmentation article: https://www.linkedin.com/pulse/machine-learning-serving-industry-herv%C3%A9-nivon