Managing the development of https://service.recap-meetings.ru (service website), collection of summarization dataset, training and fine-tuning various LLMs, diarization and ASR pipelines, optimization of inference via distillation and pruning, development of real-time transcription, activation, diarization, tone of voice analysis. Managing deployment in k8s clusters using vLLM, postgres-operator.

- Led a team of 3 engineers developing a solution for industrial Computer Vision for tracking steel slabs, training YOLOv8 and optimizing for inference in Unity, developing an algorithm for slab movement prediction and naming.-Led a team of 4 engineers developing industrial monitoring system, including collecting data from PLCs and sensors to show graphs and other information about industrial appliances, implemented alerting via Telegram, integrated InfluxDB instances into edge and server devices for seamless data transfer and aggregation. Tech stack: C# ASP.NET Core, Typescript, React, fluentd, ELK, Snap7, k8s, Kafka, InfluxDB- Led a team of 3 engineers developing a VCS for AR glasses for industrial use, implementing custom features including AEC (Acoustic Echo Cancellation) in libwebrtc for particular device. Reverse-engineered and recompiled AR glasses android app for better presentation. Tech stack: C# ASP.NET, Unity, WebRTC

Contributed to a neural net compiler, adding a MLIR dialect for conversion into SPIR-V compute shaders and launching via Vulkan, thus decreasing the inference time of some nets by ~30%. Added Constant Folding pass. Fused several kernels. Added several optimizations in MLIR dialects, decreasing memory consumption and execution time of some nets by ~5%

Built an IIOT solution for monitoring and managing sensor data in AR glasses.

Researched possible applications of Formal Verification in relation to Real-Time Operating Systems.

Reviewed Computer Architecture and Operating Systems tasks.