Hi there

🌟 Current focus

• 🔭 I’m currently working on training deep learning models and using conversational LLM/VLM models.
• 🌱 I’m currently learning more efficient coding on Leetcode.
• 👯 I’m looking to collaborate on deep learning, computer vision, LLMs, multi-modal learning.
• 📫 How to reach me: medium, linkedin.

[0.1.3] - 2024-08-28

📚 Publications

• How to share ML experiments and code (Part I, Part II)
• LLM evaluation for RAG applications
• 2023 highlights of computer vision progress
• Others: building models without labels, GPU workflows in deep learning, tf-idf + NB for text classification

💻 Open-Source Contributions

• haystack contributions: fix union types, rest api tags, ssl disable, ollama fix
• fiftyone contributions: yolov8 track ids, draw labels on video

🚀 Projects

• Real-time video analytics
• LLM-RAG service
• Structured text parsing with LLM and OCR for whatsapp bot

[0.1.2] - 2023-12-24

Publications

• Practical hyper-parameter tuning for deep learning
• Few tips to fine-tune SAM
• Others: zero-training object detection, telegram ml hackathon solution, selecting video frames for labelling

Contributions to open-source

• HOTA calculation for tracking evaluation with py-motmetrics
• fiftyone contributions: browser selection, feature extractor enhacement, dataset export tags, sam to model zoo
• tiny ultralytics contribution

Projects

• 2nd place telegram mini-app hackathon solution
• Telegram ML hackathon solution repo and a huggingface demo.

[0.1.1] - 2023-05-10

Publications

• Comparison of object detection method for SAR
• How to create reproducible data prep pipelines
• My humble manifest on keeping a Trainlog for machine learning experiments

Contributions to open-source

• Segment-anything model integration to segmentation-models-pytorch
• Segment-anything model integration to fiftyone zoo
• small hf-accelerate fix, small enhancement to segment-anything, dice score computation in fiftyone

[0.1.0] - 2023-02-03

Publications

• How to train deep nets on a remote gpu with a single git commit
• How to optimize deep learning models deployed with triton-inference-server

Contributions to open-source:

• Label studio integration in FiftyOne
• CLIP model integration in FiftyOne
• smaller contributions: triton-mlflow-plugin fix, count exact number of frames, cvat custom name, mlflow onnx flavor docs, import from/export to yolov5 format with confidence

Some projects I've worked on:

• Computer vision data prep and model training and deployment from cli. Docs are yet to come.
• UI page to set a training schedule reminder. See a webpage or try in a Telegram bot.
• Data clustering hackathon - 4th place, hackathon description.
• Telegram bot to reply with video quotes from favourite movies.
• Blog in Russian.
• Deeplearning.ai GAN specialization homework.
• Winner of epilepsy classification hackathon.