Alexander Lytchiér

Building the AI-native platform for optimizing ML inference on edge hardware such as NVIDIA Orin and Qualcomm, powered by intelligent orchestration, AI agents and production-grade operations.

Manage all software engineering challenges and run both the MLOps and performance teams. 𝐌𝐋𝐎𝐩𝐬 𝐓𝐞𝐚𝐦 - About: For a large research team, a major bottleneck with deep-learning-based experimentation is allowing for rapid experimentation while removing the software and hardware headaches typically associated with distributed GPU compute, a large software stack and code complexity. - 𝐈 𝐥𝐞𝐚𝐝 𝐚 𝐭𝐞𝐚𝐦 building novel methods tackling this new MLOps challenge, by abstracting away these layers, drastically improving research efficiency, allowing for easy management of all our infrastructure, while simplifying the complexities of deep-learning-based research. 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 & 𝐎𝐩𝐭𝐢𝐦𝐢𝐬𝐚𝐭𝐢𝐨𝐧𝐬 𝐓𝐞𝐚𝐦: - About: Deploying deep-learning-based compression algorithms on edge devices is challenging, far more so than any other kind of machine learning task, due to the complexities required for entropy coding, custom operations, memory management, large resolutions, etc. - 𝐈 𝐥𝐞𝐚𝐝 𝐚 𝐭𝐞𝐚𝐦 of experts in not only traditional performance optimisations, but also edge-device deployment of deep-learning algorithms. Our job is to make everything run faster and more efficiently on edge-device chipsets.

Building part of the future global image and video compression pipeline. Main research topic is perceptually guided learnt image and video compression. I study questions such as, how do you train lossy compression networks to be optimised for human observers? How do you model density ratios of images to compress in a manner guided by human perception?

Designed and developed a number of solutions with a multifaceted technology stack covering C++, C, Python, Java, and Golang, with a focus on mobile– and internet-based communications. - DeepVoice, a deep learning-based call quality estimator based on voice quality data acquired by the company. Insights from the model’s estimates provided a global overview of the company's internet-based call solutions, highlighting areas of infrastructure deficits or temporary issues. - Facilitated companywide adoption of TURN servers through a number of Java backend servers, C++ cross-platform library components and development of an in-house TURN solution in C. Allowing for improved call quality, routing redundancy, and better global reliability. - Identified a bottleneck for production Hadoop clusters running HBase with Apache Phoenix with a solution that improved the performance by close to two orders of magnitude. - Various improvements and bug-fixes in the mobile signaling & media engine covering VoIP, SIP and ICE-based problems along with network components such as SBCs and NGNs. - Automation of internal customer and service provisioning procedures with Golang based backend components, following Agile methodology in a team of frontend, QA and product managers.

Provided constant UX feedback and solutions, performed extensive research tasks about possible future projects, lead composer of the UDD for a new smartphone application - Continuous UX audit of the iOS & Android design-team documents of a new smartphone app; provided sound, logical solutions (in terms of flow documents) that in some cases drastically altered the behavior of the application - Extensive competitor comparison of SDK & API features (Twilio, Nexmo, Plivo, Sinch), such that the SDK project managers were fully aware of the market competitors’ state – driving further service improvements - Feasibility reports and development of internal proposals (such as an O2O expansion), reporting directly to the COO - Wrote the great majority of the User-Driven Development (UDD) documentation in a few weeks for a new smartphone application (over 30 different sections), allowing the mobile developers to quickly start development

- 80-page CEE market overview, with emphasis on Poland, Czech Republic, and Romania. Macro-analysis, including M&A, ECM, Equity market, IPO analysis of recent years. To study the viability of possible expansions into regions of CEE – first to research the area - Poland & CEE mapping for IBD bulge bracket and large local banks. Mapping hundreds of individuals using various sources and established internal databases, requiring extensive cross-referencing and precise detail

Analysis of internal mobile software, that of several competitors, and statistical analysis of user database presented in the form of a 17,000-word report and 10-minute oral presentation for 12 senior staff

Bi-weekly lead a group or class of students through numerical and problem-solving classes in physics and mathematics, in addition to a wide range of weekly subjects ranging from geography and architecture to paleontology

Final Project: A Computational Framework for Haptic Exploration and A Study of Haptic Tomography Using Self-Organizing Maps