Konstantin Zakharov Email and Phone Number

• Developed and trained diffusion model for game assets generation that was ∼30% better (FID metric) and∼3x faster than Stable Diffusion on this task• Researched approaches for fast and cheap fine tuning of diffusion models• Created APIs to access developed models using FastAPI and deployed them using Docker• Developed an animation generation model for 2D game characters• Researched 3D generation models, with technologies including: NERF, 3D diffusion models, 3D VAE• Present the research findings and developed models to Chinese management and on conferences

• Worked on super-resolution models for real-time applications on a hardware with limited compute power (smartphones, smart TVs, etc.)• Developed an upscaling model that was ~5% better (PSNR metric) and ~15x faster than SOTA super-resolution models• Suggested and integrated MLOps system for experiment tracking (ClearML)

• Developed a face recognition system for a small store• Trained a model for getting face embeddings that was ∼40% better (mAP metric) than off the shelf models• Used a HNSW algorithm to optimize storage, retrieval and comparison between face embeddings

• Developed parsers for dynamic sites and integrated it with a larger aggregator system