● Research about the techniques to make RAG quality of the product better. ● Check of the vendor providers prices of the hardware for running application in production. ● Preparing datasets for custom LLM finetuning. ● Prompt engineering techniques usage to make sure the input prompt covers the product updates. ● Helping to onboard new customers to the platform. ● Help to HR team to make clear HR policy documents.

● Creation of pre-sales materials with a focus on the technological aspects of the solution. ● Doing technical calls with the Clients with AI/ML tools suggestions. ● Creation of internal documentation about the use of AI/ML tools in the development process. ● Consulting and knowledge sharing sessions conducting. ● Conducting technical engineering interviews with the approved candidates. ● Testing third-parties AI/ML solutions before the production stage. ● Hiring people to the team. ● Built a recommendation engine and classifier model for the Client, Azure based solution. ● Made system design documents for a Client and for internal AI/ML project. ● Fine-tuned and updated Microsoft PHI-3 LLM on custom datasets with the use of Autotrain library on Hugginface.

● Making a Docker image to run Core library inside within the Openvino dependencies requirements. ● Preparing validation datasets for Detection and Tracking algorithms in MOT format. ● Preparing training datasets from customer’s data for custom Detection and Classification use cases, and training and validating models on them. ● Models conversion, optimization and delivery. ● Research on a more accurate Tracking algorithms with checking them on customers’ videos. ● Core code refactoring.

● Team Leading of a team of 4 developers - delivering tasks. ● Conducting technical interviews on Computer Vision Engineer and Backend Engineer positions. ● Code refactoring. ● Code optimization. ● Checking and testing new approaches for facial parsing algorithms. ● Making data acquisition process faster.

Responsibilities and results: ● Conducting technical interviews on Data Scientist, Machine Learning Engineer, Computer Vision Engineer positions. ● Collecting data for Object Detection and Tracking tasks for running at Jetson Nano device. ● Making Knowledge Transfer sessions with short explanation of modern approaches related to business problems. ● Providing presales materials. ● Team leading a project related to faster rendering. ● Making a lot of research about suitable approaches and datasets.

● Hired people on positions of Computer Vision Engineer, Speech Recognition Engineer, Chatbots Developer. ● Conducted technical interviews on the positions listed above. ● Collected data for training purposes on some rare use cases for object detection task. ● Made GAN pipeline for making synthetic data realistic.

● Worked on data pipelines. ● Research on 3d rendering and Gans staff.

● Dataset preparation for GAN training - data download, parsing and filtering. ● GAN models research.

● Research and enhancement of current pipeline for face swap - working on making high resolution of face swap. ● Research of face detection and landmarks detection frameworks concerning their inference time on different servers.

● Implemented a sonar images generation pipeline with GANs - the use of synthetic dataset improved quality of object detection for 14%. ● Did research on underwater dehazing pipelines. ● Trained a multi-class segmentation baseline for ice ground segmentation. ● Worked on video super resolution baseline research. ● Explore and rum demos on baseline networks for 3d multi-view reconstruction.

● Trained several GAN models for response text generation in the Russian and English languages. ● Worked on optimization of inference time - current model in production works 2.5 times faster than the previous one. ● Made a model of paragraph sentence ratio with 80% AUC quality.

● Developed a pipeline based on insight face library for Face Recognition task. ● Created a full pipeline of 3d head generation from one single image. ● Creating a database of Deep Fake videos using various deep learning techniques from downloaded and splitted Youtube videos. ● Research of applications, light-weighted neural networks and frameworks for age / race / gender detection on ARM processors. ● Researching and compiling a demo of an application for gaze detection. ● Created a dataset for the European traffic signs detection task, which has been used for training detection and classifier models. ● Helping clients with prompt engineering techniques.

● Training GANs for making synthetic datasets for face identification (standard models don’t work because of camera distortion). ● Training face identification model on prepared data and implementing it to current framework. ● Training Object Detection model on satellite imagery. ● Improved quality of face identification model to 25%.

● Did Domain Adaptation with GANs to provide data of various weather conditions for training models. It has helped to improve a quality of Detection and Instance Segmentation models up to 30% in night and rain weather conditions. ● Prepared data for training Classifier model of 200 classes (over mapped existed classes and got statistics of them). ● Trained models for Instance Segmentation task ● Made internal tools to Data Science team for better obtaining data for training models 5) Kaggling a lot at free time (Data Science Bowl -Top14%, Camera Identification Challenge - Top 39%, models for WAD CVPR Challenge (segmentation).