Doston Kholdarov

- Developing Smarty, an AI Assistant, focusing on speech processing for the Uzbek language ; - Developed ImageSearch feature for SmartBazar marketplace ; - Developed Recommendation System for SmartBazar ; - Developing Smart Monitoring System for banking app of the company ;

Developed an action detection model for Nesto Shopping Mall to track employee behavior and improve staff efficiency in branches across 5 countries including Saudi Arabia, Oman, Kuwait, Bahrain, UAE ; Trained a YOLOv10b model, achieving an mAP score of 0.848, identifying staff activities with high accuracy ; Conducted error analysis on YOLOv10 documentation to minimize detection errors ; Integrated timer logic post-action detection, refining the model’s functionality, improving real-time action recognition ;

- Worked on a project predicting car prices from a dataset of about 1M+ rows, implementing rigorous data cleaning and feature engineering ; - Reduced missing data from 40% to 9% and corrected data discrepancies, improving dataset integrity for modeling ; - Utilized advanced modeling techniques, achieving an RMSE reduction from 12% to 5.5% with Random Forest, and developed a web application using StreamLit to visualize model performance metrics ;

- Performed data cleaning and preprocessing tasks on multiple data projects ; - Helped ML team to optimize classification models with hyperparameter tuning techniques ; - Developed group and individual lessons for 15+ interns in “Machine Learning” program to teach data manipulation tools and machine learning algorithms of supervised learning ; - Mentored interns in project implementation and model development, helping them to complete their first real-world machine learning project ;

- Taught Data Science, Machine Learning, Deep Learning, Mathematics, and Statistics to more than 100 students, enhancing their exam preparation and conceptual understanding ; - Leveraged provided learning materials to design tailored lessons, improving student outcomes in highly technical fields ; - Consistently broadened and solidified subject matter expertise through self-learning and teaching, contributing to student satisfaction and retention ;

Developed AI-powered voice calculator for the Uzbekistan government, achieving 90% accuracy using CNN to recognize and perform mathematical operations in the Uzbek language ; Trained CNN model with 90% accuracy after collecting and processing 2,800 voice samples across 14 classes ; Improved model accuracy to 93% by training an LSTM-based model on a 10,000-voice dataset retrieved from CommonVoice and optimized it with GridSearchCV ;

- Developed an application for the South Korean government to detect illegal traffic events using NVIDIA’s AGX Jetson Nano ; - Created a dataset of 40,000 images and implemented data augmentati on, achieving a mean Average Precision (mAP) of 0.959 on YOLOv5 model training aft er hyperparameter opti mizati on ; - Collaborated with team members to ensure model accuracy and performance; produced a demo video (https://hwaseong-agx.netlify.app/) for client presentation ;