Hi, I'm Maksim
ML Engineer at Kandinsky Lab specializing in VLM/LLM fine-tuning on multi-GPU infrastructure. First-year master's student at ITMO AI Talent Hub.
Field of interest: Computer Vision, VLM, LLM, Sign Language Recognition, Multimodal AI
Link to my English CV
- DL Engineer at Kandinsky Lab (October 2025 – Present)
- Full-finetune and LoRA tuning large-scale VLM models (Qwen MoE series) on multi-GPU/multi-node infrastructure
- DL Engineer at SBER (November 2024 – October 2025)
- Multi-agent LLM system development (LangChain, LangGraph)
- Text classification, NER, RAG for agent decision-making
- DL Researcher at SBER AI (April 2024 – November 2024)
- Satellite image segmentation (DETR, SWIN-DETR, RT-DETR, DINO)
- VLM/LLM pipelines (Qwen, LLaMA, PPLaVA)
- Memory-Augmented VLM for Sign Language Recognition (in progress)
- End-to-end VLM architecture with memory bank attention (inspired by SAM2) for SLR
- ViT encoder → Self-Attention + Cross-Attention with visual/text memory → Projector → LLM (Qwen) generation
- Evaluated on WLASL, Slovo (Russian SL), SWL-LSE (Medical SL)
- Russian Sign Language Recognition — Bachelor thesis, NUST MISIS
- 🥇 1/13 FIC Hackathon — ML model for candidate expertise evaluation | Solution | Certificate
- 🥈 2/108 E-CUP by Ozon Tech — Product matching by names, attributes & images | Solution | Certificate
- 5/30 IT Purple Hack — LLM-based QA system for Central Bank of Russia | Solution
- Finalist — Gagarin Hackathon — Telegram financial sentiment analysis | Solution
- Finalist — Tender Hack Moscow — CTE characteristics generation | Solution
- Finalist — VTB MORE.TECH 5.0 | Solution | Certificate
- Finalist — Deep Learning School Olympiad
Languages: Russian (Native), English (C1)
Contacts: Telegram | LinkedIn | maksimcrewceo@gmail.com