Aleksandra Gorokhova

Computer vision engineer and researcher

💻 Techical skills

Python,
PyTorch,opencv, PIL,timm,transformers
JIT, nvidia-dali, kornia
classic CV, VLMs, LLM
Flask, Django
linux, git, docker

👩‍💻 Work Experience

Junior CV Reseacher @ VisionLabs (Sept 2024 - Present)
Works with: classic CV, VLMs, LLMs
Development of pipeline of determination of vehicle orientation. Acceleration of algorithms.

Data Scientist @ Sberbank Cyber Security. Anti-Fraud Department. Management of AI Modeling and Development (May 2024 - Present)
Work with: pyspark, pandas, scikit-learn, hive, python, sql, catboost, lightgbm

Development of a fraud detection model for transfers to individuals entity.

Building data marts.
Writing optimized queries.
Creating new features.
Development and training of new models.
Additional retraining.

Computer Vision Researcher/Engineer @ Sberbank Robotics Center (May 2023 - March 2024)
Work with: python, pytorch, opencv, scikit-image, Pillow, CNNs, Transformers, Segformer, Docker

Inverse lithography project

Development of a full-fledged pipeline for generating synthetic data that repeats the original images from an electron microscope.
Improved the accuracy of segmentation models by using generated synthetic data.
Wrote additional software functionality on top of image noise reduction models - Blind2Unblind, Scunet.
Implemented algorithms for calculating the pure signal of an electron microscope from scientific articles.
Have written a lot of user interfaces for developed solutions.

👩‍🎓 Education

ITMO (2024 - 2026)
Artificial Intelligence, Computer Science and Computer Engineering, Master’s degree

Student of AI Talent Hub.

BMSTU (2020 - 2024)
Robotics and Complex Automation, Computer Science and Computer Engineering, Bachelor’s degree
GPA: 4.78/5

🧬 Science expirience

Final year Bachelor’s thesis (2024)
Generation of real time video subtitles based on the movement of the speaking person’s lips.
Work with: python, pytorch, pytorch-lightning, opencv, pillow, CNN, Transformers, django, vannila js, docker
Results:
- Collected, preprocessed and annotated data for training (lots of videos)
- Developed Model Architecture using Conv3D + modified EfficientNetV2 + Transformer Encoder + TCN using description in the article
- Made experiments with EfficientNetV2 parameters, sizes and with different videos languages (russian and english)
- Made experiments with classification of visemes (visual representation of phoneme).
- Developed WEB App for demo using Django and Vanilla JS
  The current version of the project is a classification of spoken words..
Scientific article “Implementation of verification methods for elements of discrete event models for robotics tasks” (2023)
Publisher: Integrated automation of design and production, BMSTU Press
Modeling of processes in a discrete event system for managing a group of mobile robots. Improving the quality of the system by verifying the model.
Report “Implementation of verification methods for elements of discrete event models for robotics tasks” (2023)
Publisher: Integrated automation of design and production 2023 KAPPA(M)-2023

🏆 Accomplishments

Won 2nd Place @ Digital Breakthrough. Season: Artificial Intelligence (May 2023)
Implementation of a model for predicting the addition of a person to friends.

FCE (2019)
English Upper-Intermediate.

💬 Languages

🇷🇺 Russian: native
🇺🇸 English: B2+ (Upper-Intermediate)