Uan is available for hire
Hire UanUan Sholanbayev
Verified Expert in Engineering
Deep Learning Engineer and Software Developer
Almaty, Almaty Province, Kazakhstan
Toptal member since October 11, 2019
Bio
After earning a degree in computer engineering from the University of California, San Diego, Uan has gained much in-depth and hands-on experience delivering services as a machine learning engineer. He is an expert in Python, deep learning, CNN, LLMs, and other AI algorithms. Uan is also an exemplary communicator and has worked in teams, as well as individually.
Portfolio
Narya AI
Python 3, Machine Learning, Natural Language Processing (NLP)...
G42
PyTorch, Python 3, Solidity, You Only Look Once (YOLO), Game Theory...
MindsLenses
Amazon Web Services (AWS), Python, PyTorch, SQL, Machine Learning...
Experience
- Deep Learning - 7 years
- Git - 7 years
- Python - 7 years
- MongoDB - 7 years
- PostgreSQL - 6 years
- Amazon EC2 - 6 years
- PyTorch - 6 years
- Computer Vision - 5 years
Availability
Part-time
Preferred Environment
GitLab, PyCharm, Ubuntu, MacOS, Amazon EC2, MongoDB, Python 3, PyTorch, You Only Look Once (YOLO)
The most amazing...
...The thing I've built is an NBA event prediction pipeline—the main product of a startup, MOCAP Analytics, which was later successfully acquired by Sportradar US.
Work Experience
Machine Learning Engineer
2023 - 2024
Narya AI
- Finetuned a 7B LLM on Amazon EC2 using LoRa, PEFT, and Hugging Face. Deployed via a 3rd-party company product.
- Developed the recommendation system, relevancy filtering, summarization, and NER using GPT and finetuned open-source LLM.
- Deployed on Heroku Solidity similarity search and gas-optimization APIs based on generative AI and vector databases.
- Extracted the issues dataset from PDF reports using LLM and uploaded it to Supabase vector databases.
- Automated Solidity hacking bug detection using LLM and RAG.
Technologies: Python 3, Machine Learning, Natural Language Processing (NLP), Large Language Models (LLMs), OpenAI, Agile, Deep Learning, Artificial Intelligence (AI), Amazon EC2, Amazon S3 (AWS S3), Python, PyTorch, Git, PostgreSQL, SQL, Agile Software Development, Neural Networks, Random Forests, Tf-idf, Scikit-learn, Slack, Sublime Text 3, MacOS, Generative Pre-trained Transformers (GPT), Sublime Text, Generative Artificial Intelligence (GenAI), Data Science, Data Structures, Object-oriented Programming (OOP), Data Versioning, K-nearest Neighbors (KNN), Algorithms, Data Mining, Object Tracking, Statistical Modeling
Machine Learning Engineer
2021 - 2023
G42
- Delivered a computer vision project that automatically reads ID document scans.
- Developed computer vision software to detect road traffic violations.
- Built a PCR registration chatbot, later used as the basis to create other services.
Technologies: PyTorch, Python 3, Solidity, You Only Look Once (YOLO), Game Theory, Object Tracking, Chatbots, Rasa NLU, Moralis, Ubuntu, Microsoft Excel, Jira, Agile Software Development, Machine Learning, Random Forests, Linear Regression, Logistic Regression, PyCharm, XGBoost, Scikit-learn, Sublime Text 3, Sublime Text, Video Processing, Object Detection, Agile, Deep Learning, Artificial Intelligence (AI), Computer Vision, Python, Git, PostgreSQL, GitLab, Neural Networks, Slack, MacOS, Convolutional Neural Networks (CNNs), Image Processing, Data Science, Data Structures, Object-oriented Programming (OOP), K-nearest Neighbors (KNN), Algorithms, Data Mining, OpenCV, Statistical Modeling
Computer Vision Developer
2019 - 2020
MindsLenses
- Created a pipeline that processes everyday movies collected from RPI cameras. The processing includes several deep learning algorithms that detect human emotions associated with product interactions being handled in the store.
- Trained deep learning models for object detection and classification. Used state-of-the-art frameworks to produce the best-quality predictions.
- Managed stable process of videos from raspberry pies to CV inference on the AWS cloud.
Technologies: Amazon Web Services (AWS), Python, PyTorch, SQL, Machine Learning, Object Detection, Agile, Artificial Intelligence (AI), Computer Vision, Amazon EC2, Amazon S3 (AWS S3), Git, PostgreSQL, Agile Software Development, GitLab, Neural Networks, K-means Clustering, Slack, Sublime Text 3, MacOS, Convolutional Neural Networks (CNNs), Image Processing, Video Processing, Sublime Text, Data Science, Data Structures, Object-oriented Programming (OOP), K-nearest Neighbors (KNN), Algorithms, Data Mining, Object Tracking, OpenCV, Statistical Modeling
Machine Learning Engineer
2017 - 2019
Sportradar US
- Created a neural network for the NFL to generate drive outcome predictions for further use by betting companies.
- Built NBA shot probability neural network models using PyTorch for fan experience enhancement.
- Identified NBA events on games using tracking features with XGBoost.
- Implemented an NBA data pipeline from different data sources.
- Created an NBA similarity model for draft players for a fan-enhanced experience.
Technologies: Python, Machine Learning, Agile, Deep Learning, Artificial Intelligence (AI), Amazon EC2, Amazon S3 (AWS S3), PyTorch, Git, MongoDB, SQL, Agile Software Development, GitLab, Neural Networks, Random Forests, Linear Regression, Logistic Regression, Tf-idf, K-means Clustering, XGBoost, Scikit-learn, Slack, Robo 3T (Robomongo), Sublime Text 3, MacOS, Natural Language Processing (NLP), Sublime Text, Data Science, Data Structures, Object-oriented Programming (OOP), Data Versioning, K-nearest Neighbors (KNN), Algorithms, Data Mining, Statistical Modeling
Data Scientist
2016 - 2017
Alem Research
- Scraped 10 million company front pages from a social network with Java.
- Created a sentiment analysis neural network model using TensorFlow to label various English texts (including articles, comments, and more) as positive, negative, or neutral.
- Self-retrained a text topic relevance model on the AWS cloud.
Technologies: Java, Python, Machine Learning, Natural Language Processing (NLP), Deep Learning, Artificial Intelligence (AI), PyTorch, Git, TensorFlow, Agile Software Development, GitLab, Neural Networks, Random Forests, Tf-idf, K-means Clustering, XGBoost, Slack, Sublime Text 3, Sublime Text, Data Science, Data Structures, Object-oriented Programming (OOP), K-nearest Neighbors (KNN), Algorithms, Data Mining, Statistical Modeling
Experience
Cloud Video Processing with ML
1. Led the video processing pipeline for a retail analytics project, focusing on customer interaction with products.
2. Devised classifiers for accurate shoe detection and customer demographic analysis via video footage.
3. Innovated cloud-based processing workflows on AWS to handle video analytics and database management.
2. Devised classifiers for accurate shoe detection and customer demographic analysis via video footage.
3. Innovated cloud-based processing workflows on AWS to handle video analytics and database management.
Traffic Violation Detection
Engineered a machine learning pipeline for real-time video analysis, identifying traffic violations using YOLO, the SORT algorithm, and Fast API. Trained a separate "taxi" and "trash" class with merged public datasets. Created a custom framework to apply all models and detect traffic violations with model+xy coordinates-based rules.
NFL Drive Outcome Prediction
I implemented a Pytorch-based framework to train neural network models for NFL drive outcome predictions. The predictions are used by betting companies to generate coefficients.
1. Developed a robust data framework to facilitate rapid model training for national college football analytics.
2. Executed predictive drive outcome models for the NFL, utilizing neural networks to derive betting coefficients.
1. Developed a robust data framework to facilitate rapid model training for national college football analytics.
2. Executed predictive drive outcome models for the NFL, utilizing neural networks to derive betting coefficients.
Neural Network Distilling Knowledge
https://github.com/usholanb/Experiments-with-Distilling-Knowledge.gitOne of the hottest topics in the field of neural networks is knowledge distillation. Data scientists are curious about how compacted a neural net can become while still performing at a similar quality level.
The approach is to use the layer before the last (output layer) and use it as an output layer for a smaller neural net (with less hidden layers).
In this project, I used ten classes of ImageNet (around 1,000 pictures each) and trained two such neural nets. The results were the large model was 68% top 1. The smaller model (trained itself) was 56% top 1, and the distilled model was 56.5% top 1.
The approach is to use the layer before the last (output layer) and use it as an output layer for a smaller neural net (with less hidden layers).
In this project, I used ten classes of ImageNet (around 1,000 pictures each) and trained two such neural nets. The results were the large model was 68% top 1. The smaller model (trained itself) was 56% top 1, and the distilled model was 56.5% top 1.
Article Ranking
Spearheaded the development of a text relevancy ranking algorithm using BERT embeddings and LR classification with cross-encoders for enhanced ranking precision. Deployed on Heroku with FastAPI. Utilized LLMs on top to further improve relevancy ranking.
Education
2012 - 2016
Bachelor of Science Degree in Computer Engineering
University of California, San Diego - San Diego, CA, USA
Certifications
SEPTEMBER 2017 - PRESENT
Machine Learning
Stanford University | via Coursera
Skills
Libraries/APIs
OpenCV, PyTorch, Scikit-learn, XGBoost, TensorFlow, Rasa NLU, PiLLoW, Matplotlib
Tools
Sublime Text, Git, Jira, Slack, Robo 3T (Robomongo), PyCharm, Sublime Text 3, Microsoft Excel, MATLAB, You Only Look Once (YOLO), GitLab, PyInstaller
Languages
Python, Python 3, Java, SQL, C++, Scala, Solidity
Paradigms
Object-oriented Programming (OOP), Agile, Agile Software Development
Platforms
Amazon EC2, Ubuntu, MacOS, Amazon Web Services (AWS)
Storage
Amazon S3 (AWS S3), MongoDB, PostgreSQL
Frameworks
Caffe
Other
Artificial Intelligence (AI), Convolutional Neural Networks (CNNs), Image Processing, Machine Learning, Video Processing, Neural Networks, Statistical Modeling, Deep Learning, Random Forests, Linear Regression, Logistic Regression, Data Versioning, Tf-idf, K-means Clustering, K-nearest Neighbors (KNN), Computer Vision, Object Tracking, Data Structures, Algorithms, Data Mining, Data Science, Natural Language Processing (NLP), Chatbots, Moralis, Generative Pre-trained Transformers (GPT), Generative Artificial Intelligence (GenAI), Large Language Models (LLMs), OpenAI, Object Detection, Game Theory, Object Recognition, Tkinter, Image Classification, Large Data Sets, FastAPI, Embeddings from Language Models (ELMo)
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring