Zing Coach™

Senior R&D Engineer - Computer Vision

Aiuta

• Implemented a Transformer model for outfit scoring and
generation, enhancing AI stylist capabilities.
• Performed advanced prompt engineering techniques for AI stylist
bot, improving user interaction and satisfaction.
• Improved a LLM chatbot using RAG for banking services

Senior Data Scientist

Checklens

-Developed SOTA detection and ReID models for self-checkout loss
prevention, leveraging knowledge of business processes.
Optimized model performance for real-time using knowledge
distillation and TensorRT, resulting in a 15% decrease in
loss-related incidents and improved customer satisfaction.

- Initiated and led the implementation of ClearML as the primary
MLOps platform for managing datasets, pipelines, and
experiments.

Senior Computer Vision Researcher

Lamoda Group

– Leading the Computer Vision division at the company.
– Explored state-of-the-art DL in fashion. Trained a RegNet on multiclass fashion tag classification and self-supervised contrastive loss for Similar Fashion Recommendations, Automatic Tagging, Visual Search. 
– Implemented a Transformer-based network for Automated Outfit Generation. (PyTorch-Lightning, Hydra configs, ClearML)
– Created a large-scale inference pipeline with ONNX and Airflow

Optia.ai 

– Managed a team of 3 people to make the product be enterprise-ready. Improved team efficiency twice with data-driven agile and ICE score prioritization.
– Trained a transformer model to classify text sentiment from Twitter and Instagram posts (PyTorch, 90.2% precision on an inner dataset).
–  Implemented a texts search pipeline using Sentence Transformers embeddings with FAISS and Prefect.
– Engineered a reproducible data pipeline for social media analysis using Prefect. 2x faster release rolling.

Dbrain

- Implemented and trained RetinaNet, CenterNet, and U-Net with PyTorch to perform state-of-the-art OCR on documents and words recognition.
- Improved the mAP metric from 0.66 to 0.81 for symbol detection using RetinaNet. 
- Developed a tool for generating synthetic OCR dataset in Python
- Develop a pipeline to analyze cattle behavior. This involves identifying individual animals (CenterNet detector + Sphere Loss; PyTorch) and understanding their actions.
- Deployed DL models into production using Docker, PostgreSQL and FastAPI.

Data Scientist

Kinoplan

- Trained a WideResNet age-gender classification model for cinema ticket terminals using Keras.
- Engineered a tracking system for movie theater bars to prevent stealing goods (YOLOv3 detector + SORT tracker; Keras).

Developed software for cinemas: automation of scheduling movie theaters sessions; a chat system for communication between distributors and cinemas  (Scala, MongoDB).

Software Engineer

Deep Learning engineer with 6+ years of expertise in CV and NLP, committed to MLOps and ML Reproducibility.

My resume in PDF: https://drive.google.com/file/d/1bvXeqwM7mhxpHBcWywlYWvP20oDieErb/view?usp=sharing

Taplio

Roman Tezikov's Linkedin Analytics

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Roman Tezikov's Linkedin Analytics

Want to drive more opportunities from LinkedIn?

Roman Tezikov's Linkedin Analytics