Olga Ivanova

Developed end-to-end computer vision solutions including an InsightFace-based face recognition pipeline, a single-image 3D head model pipeline, and a deepfake video database from YouTube. Deployed an Azure AI Platform AI Search Agent (Azure OpenAI, Form Recognizer, Cognitive Search, vector databases) and optimized/maintained tracking, alignment, and dataset pipelines for production use.

Built an agent that interacts with a Power BI dashboard to analyze its data, and supported development of an object detection pipeline. Refactored code and evaluated technology options for planned development work.

Improved RAG quality by researching techniques, prepared datasets for custom LLM fine-tuning, and applied prompt engineering to ensure prompts reflect product updates. Supported production readiness through hardware cost evaluation and helped onboard new customers to the platform.

Built an Azure-based recommendation engine and classifier model, and produced client/internal system design documentation and pre-sales technical materials. Fine-tuned and updated Microsoft Phi-3 on custom datasets using Autotrain (Hugging Face) and supported delivery through technical calls, third-party solution testing, interviews, and hiring.

Developed a Docker image to run OpenVINO Core with required dependencies and prepared MOT-format validation datasets for detection and tracking. Trained, converted, optimized, and delivered custom detection/tracking models from customer data and refactored the core codebase.

Led a team of 4 developers, covering task delivery, technical interviews, and engineering reviews. Refactored and optimized code, researched facial parsing approaches, and sped up the data acquisition process.

Collected data for object detection and tracking pipelines targeting Jetson Nano deployment and led technical interviews for DS/ML/CV roles. Provided knowledge-transfer and pre-sales materials and supported a team effort to speed up rendering by researching approaches and datasets.

Implemented a sonar-image generation pipeline using GANs; the synthetic dataset improved object detection quality by 14%. Researched underwater dehazing, trained a multi-class segmentation baseline for ice ground segmentation, explored video super-resolution, and investigated 3D multi-view reconstruction baselines.

Hired and supported onboarding for Computer Vision, Speech Recognition, and chatbot development roles through technical interviews. Collected data for rare object-detection use cases and built a GAN pipeline to generate realistic synthetic data.

Worked on data pipelines and researched 3D rendering approaches and GAN-based methods. Supported development through iterative experiments and integration of research results into workflows.

Prepared datasets for GAN training by downloading, parsing, and filtering data, and conducted GAN model research for vision use cases. Delivered reusable data-preparation pipelines to support training iterations.

Enhanced a face-swap pipeline by researching and improving output resolution and quality. Researched face detection and landmark frameworks and benchmarked their inference time across different servers.

Trained multiple GAN models for response text generation in Russian and English and optimized inference time, delivering a production model 2.5× faster than the previous one. Built a paragraph-sentence ratio model achieving 80% AUC quality.

Trained GANs to generate synthetic datasets for face identification where standard models struggled due to camera distortion, and trained a face identification model integrated into the existing framework. Trained an object detection model for satellite imagery and improved face identification quality by 25%.

Performed GAN-based domain adaptation to improve detection and instance segmentation quality by up to 30% in night and rain conditions. Prepared data for a 200-class classifier, trained instance segmentation models, built internal tools for faster training-data acquisition, and achieved top Kaggle placements (Top14%, Top39%).

Grade: 8.0

Studied at the Faculty of International Relations and graduated as an economist-analyst with an average grade of 8.0.

Completed secondary education at Gymnasium №7 in Minsk.