Abay is available for hire

Abay Bektursun

Verified Expert in Engineering

Artificial Intelligence Engineer and Developer

Austin, TX, United States

Toptal member since July 27, 2022

Expertise

Dashboard Machine Learning Deep Learning Artificial Intelligence Neural Network Data Science Back-end Developers Analytics Development Data Scraping

Bio

Abay is an AI engineer and tech leader specializing in computer vision and scalable AI systems. At Apple, he led the development of a global, multi-million-dollar computer vision product. He drove the technical vision as CTO and co-founder of Gridlines (raised $1 million). At Docme.ai, he created models measuring vital signs via an iPhone camera. At Copycopter.ai, he developed scalable AI for video and image generation. Abay leads a community of over 3,000 AI enthusiasts.

Portfolio

AbstractAI

AI Design, Generative Pre-trained Transformers (GPT), Stable Diffusion...

Eagle Eye Networks

C, Machine Learning, Deep Learning, Computer Vision, Linux, Python 3...

Apple

TensorFlow, Deep Learning, Computer Vision, System Design, Python 3...

Experience

Python 3 - 8 years
Deep Learning - 7 years
Machine Learning - 7 years
Computer Vision - 5 years
TensorFlow - 4 years
C++ - 3 years
PyTorch - 3 years
Computational Neuroscience - 1 year

Availability

Full-time

Preferred Environment

Linux, Deep Learning, Artificial Intelligence (AI), PyTorch, Python 3, Hugging Face

The most amazing...

...project I've led was computer vision product at Apple.

Work Experience

Autonomous AI Expert

2022 - PRESENT

AbstractAI

Played a key role in establishing a finance AI startup, which successfully raised over $1 million in capital: https://gridlinesapp.com/.
Helped a startup build computer vision capabilities that they were able to sell to the Japanese government: https://www.cropwatch.io/.
Built a SOTA model for measuring health signals via iPhone camera: https://www.docme.ai/. Managed a team of four.
Built an end-to-end vision system for a fashion startup.
Built a scalable video and image generation system for https://copycopter.ai/.

Technologies: AI Design, Generative Pre-trained Transformers (GPT), Stable Diffusion, Fine-tuning, LoRa, OpenAI GPT-4 API, Generative Pre-trained Transformer 3 (GPT-3), Workshop Facilitation, OpenAI GPT-3 API, Fairseq, ChatGPT, PyTorch Lightning, Data Scraping, Statistical Methods, Statistical Data Analysis, Statistical Analysis, Language Models, CSS, HTML, Distributed Computing, OpenAI, Llama 2, Falcon, PEFT, BERT, LSTM, Software Architecture, Chatbots, Dashboards, Speech to Text, Google Speech-to-Text API, CTO, Retrieval-augmented Generation (RAG), Serverless, Model Tuning, AWS Lambda, LangChain, Natural Language Processing (NLP), ChatGPT Prompts, ChatGPT API, Prompt Engineering, Open-source LLMs, Rust, Convolutional Neural Networks (CNNs), Object Recognition, OpenAI API, Llama, Gemini, Gemini API, Hugging Face Transformers, Edge AI, NVIDIA TensorRT, Quantization, model quantization, Text to Speech (TTS), Hugging Face, Pattern Recognition

Computer Vision Engineer

2020 - 2022

Eagle Eye Networks

Developed embedded vision features deployed to tens of thousands of cameras worldwide.
Prototyped state-of-the-art deep learning methods for surveillance computer vision by harnessing large amounts of surveillance video.
Created prototypes with various edge accelerators for computer vision.

Technologies: C, Machine Learning, Deep Learning, Computer Vision, Linux, Python 3, TensorFlow, C++, Leadership, Project Leadership, Team Leadership, Python, NumPy, JSON, CSV, Artificial Intelligence (AI), Machine Learning Operations (MLOps), Image Processing, Neural Networks, Cloud, Machine Vision, Data Engineering, Data Reporting, Data Analytics, Artificial Neural Networks (ANN), Scripting, Deep Neural Networks (DNNs), PyTorch, Software Engineering, Cloud Services, DevOps, Pandas, SQL, Linear Regression, Clustering, Visualization Tools, Docker, Google Cloud Platform (GCP), Modeling, Data Mining, Back-end, Distributed Systems, GitHub, Back-end Development, Facial Recognition, Google Cloud, REST APIs, Scikit-learn, Keras, Large Language Models (LLMs), Generative Research, AI Design, Internet of Things (IoT), Analytics, Computer Vision Algorithms, OpenCV, Data Analysis, Jupyter, Git, CCTV, Real-time Data, Hardware, Architecture, Hugging Face, Fine-tuning, Statistical Data Analysis, Statistical Analysis, CSS, HTML, LSTM, Software Architecture, Dashboards, Serverless, Model Tuning, Image Recognition, Convolutional Neural Networks (CNNs), Object Recognition, Hugging Face Transformers, Edge AI, NVIDIA TensorRT, Quantization, model quantization, Pattern Recognition

Computer Vision Engineer

2019 - 2020

Apple

Developed the vision system that detects people's presence in Apple stores.
Led the team that developed a computer vision system for Apple store analytics.
Applied ideas from an academic research paper to a real-world product.

Technologies: TensorFlow, Deep Learning, Computer Vision, System Design, Python 3, Object Detection, Object Tracking, Machine Learning, Leadership, Project Leadership, Team Leadership, Python, NumPy, JSON, CSV, Artificial Intelligence (AI), Data Modeling, Machine Learning Operations (MLOps), Image Processing, Neural Networks, Cloud, Machine Vision, Data Engineering, Data Reporting, Data Analytics, Artificial Neural Networks (ANN), Scripting, Deep Neural Networks (DNNs), PyTorch, Software Engineering, Cloud Services, DevOps, Generative Adversarial Networks (GANs), Pandas, SQL, Pytest, Linear Regression, Clustering, Visualization Tools, Docker, Modeling, Data Mining, Back-end, Distributed Systems, GitHub, Java, Back-end Development, Facial Recognition, Data Pipelines, TypeScript, Google Cloud, REST APIs, Scikit-learn, Keras, Generative Research, AI Design, Analytics, eCommerce, Signal Processing, Computer Vision Algorithms, OpenCV, Data Analysis, Jupyter, Git, CCTV, Real-time Data, Hardware, Architecture, Hugging Face, Fine-tuning, Statistical Data Analysis, Statistical Analysis, CSS, HTML, BERT, LSTM, Software Architecture, Dashboards, Serverless, Model Tuning, Image Recognition, Convolutional Neural Networks (CNNs), Object Recognition, Edge AI, NVIDIA TensorRT, Quantization, model quantization, Pattern Recognition

Machine Learning Developer

2016 - 2019

Hewlett Packard Enterprise

Joined the company as an intern and was recognized as one of the top three interns.
Led a development team for an entirely automated financial department. Reported to the CEO and saved the company $3 million.
Took leadership roles outside everyday work. Led employee volunteering programs, organized hackathons, and taught technical classes on Linux and machine learning.
Participated in NLP projects, summarizing and classifying company reviews to improve branding and analyzing employee survey text to improve the company culture.

Technologies: Hadoop, Python 3, Machine Learning, Data Science, Data Visualization, Tableau, TensorFlow, Deep Learning, C++, Leadership, Project Leadership, Team Leadership, Python, NumPy, JSON, CSV, Word2Vec, Artificial Intelligence (AI), Data Modeling, Machine Learning Operations (MLOps), Robot Operating System (ROS), Image Processing, Forecasting, Neural Networks, Cloud, Machine Vision, Data Engineering, Data Reporting, Data Analytics, Artificial Neural Networks (ANN), Scripting, Automation, Automated Data Flows, Deep Neural Networks (DNNs), PyTorch, Software Engineering, Cloud Services, DevOps, Text Mining, Pandas, SQL, ETL, Linear Regression, Clustering, Visualization Tools, Amazon Web Services (AWS), Docker, Google Cloud Platform (GCP), Modeling, Predictive Modeling, Predictive Analytics, Data Mining, Back-end, Distributed Systems, GitHub, Java, Back-end Development, Facial Recognition, Web Scraping, Data Pipelines, Google Cloud, REST APIs, Big Data, Scikit-learn, Keras, AI Design, Internet of Things (IoT), Analytics, Computer Vision Algorithms, OpenCV, Amazon S3 (AWS S3), Data Analysis, Jupyter, Git, Reinforcement Learning, Real-time Data, Hardware, Architecture, Workshop Facilitation, Data Scraping, Statistical Data Analysis, Statistical Analysis, Language Models, CSS, HTML, Distributed Computing, Financial Modeling, BERT, LSTM, Software Architecture, Chatbots, Dashboards, Model Tuning, Image Recognition, Convolutional Neural Networks (CNNs), Object Recognition, NVIDIA TensorRT, Quantization, model quantization, Pattern Recognition

Software Engineer Intern

2015 - 2016

Centene

Developed and maintained a documentation website, both its front-end and back-end work. Wrote scripts to process and parse EDI files.
Ran routine jobs and processed health insurance claims. Automated manually run jobs and reports.
Produced ad-hoc and scheduled reports for different departments. Helped vendors resolve issues and support third-party software.

Technologies: Python 3, Oracle, Databases, MongoDB, Electronic Data Interchange (EDI), Neural Networks, Cloud, Machine Vision, Data Engineering, Data Reporting, Data Analytics, Artificial Neural Networks (ANN), Scripting, Automation, Automated Data Flows, Deep Neural Networks (DNNs), PyTorch, Software Engineering, Cloud Services, DevOps, Django, Pandas, SQL, ETL, Linear Regression, Clustering, Visualization Tools, Modeling, Predictive Modeling, Predictive Analytics, Data Mining, Back-end, GitHub, Back-end Development, Web Scraping, REST APIs, Scikit-learn, Keras, Internet of Things (IoT), Analytics, OpenCV, Data Analysis, Jupyter, Git, Statistical Data Analysis, Statistical Analysis, CSS, HTML, Software Architecture, Dashboards, Model Tuning, Image Recognition, Convolutional Neural Networks (CNNs), Object Recognition, Pattern Recognition

Experience

AI for Wealth Management

https://elara.tech/

A powerful AI-driven workflow automation solution tailored for wealth managers integrating CRM and custodial platforms. Bootstrapped from the ground up, and now actively raising funds for scaling and growth.

Why Does Batch Normalization Work?

A theoretical and experimental exposition on Batch normalization that explains the real reason why it works so well. The ML community believes Batch Norm improves optimization by reducing internal covariate shift (ICS). As I show, ICS has little to no effect on optimization.

Built a Community of Three Thousand People

https://www.meetup.com/Austin-Deep-Learning/

Austin Deep Learning is the largest deep learning community in Texas. We invite talks from machine learners and data scientists applying deep learning to solve problems, with tutorials and lessons learned. Talks are open to all deep learning frameworks, such as TensorFlow, Keras, PyTorch, and others.

Large Language Models | Alignment Experiment

https://www.linkedin.com/feed/update/urn:li:activity:7072309610354728960/

An experiment on aligning large language models (LLMs) was conducted for research during the OSS LLMs workshop. The main hypothesis, proposed by David Ha, aimed to investigate whether a pronounced moral bias could potentially compromise the core effectiveness of LLMs.

The participants were invited to design three tasks with varying difficulty levels – elementary, intermediate, and advanced – tailored explicitly for LLMs. The models used for this experiment included Vicuna-13B, Vicuna-13B Uncensored, and Vicuna-7B, which served as a baseline for comparison. The participants assessed and rated the performance of each model based on their respective tasks.

The central focus was to examine the impact of intense alignment bias on the overall efficacy of LLMs. The findings of this study provided substantial evidence to support the hypothesis. The Vicuna-13B Uncensored model, trained on an augmented dataset with fewer moral constraints, achieved an average score of 5.75 out of 10, whereas the censored model secured an average of 3.95. This observation could be attributed to the tendency for stronger alignment to encourage deeper mode-seeking within the model distribution.

Raw CUDA Implementation of Gaussian Splatting

https://github.com/abaybektursun/gauss_splats

Lately, I've been diving into computer graphics and GPU programming. Thanks to the GPUMODE Lectures and the support of LLM assistants, the learning curve has been much more manageable. I'm documenting my project journey and sharing the insights I've gained so far in the following resources:

Project Overview: Brook-ankle-f27.notion.site/Gaussian-Splats-in-CUDA 16ceb624d4e980c78338dd24954d0c44
Source Code: Github.com/abaybektursun/gauss_splats

Feel free to check them out!

Education

2014 - 2017

Bachelor's Degree in Computer Science

University of Central Arkansas - Conway, AR, USA

Certifications

SEPTEMBER 2022 - PRESENT

Inferential Statistical Analysis

University of Michigan, via Coursera

SEPTEMBER 2022 - PRESENT

AWS Machine Learning

AWS, via Coursera

MAY 2018 - PRESENT

Deep Learning Specialization

DeepLearning.AI, via Coursera

NOVEMBER 2017 - PRESENT

Machine Learning Specialization

Stanford University, via Coursera

NOVEMBER 2016 - PRESENT

The Arduino Platform and C Programming

Coursera

Skills

Libraries/APIs

NumPy, PyTorch, REST APIs, Keras, OpenCV, LSTM, Google Speech-to-Text API, OpenAI API, Hugging Face Transformers, TensorFlow, Pandas, Scikit-learn, PyTorch Lightning, Node.js

Tools

GitHub, Jupyter, Git, NVIDIA Jetson, NVIDIA Nsight Systems, Tableau, Vendor Independent Messaging (VIM), Scikit-image, Pytest, ChatGPT, Google AI Platform

Languages

Python 3, Python, SQL, CSS, HTML, Falcon, JavaScript, Java, TypeScript, C, C++, Rust, Embedded C

Paradigms

Automation, ETL, Parallel Programming, DevOps, Web App Design, Distributed Computing

Platforms

Visual Studio Code (VS Code), Linux, Amazon Web Services (AWS), Docker, Google Cloud Platform (GCP), AWS Lambda, MacOS, Oracle, Arduino, AWS IoT, NVIDIA CUDA

Storage

JSON, Google Cloud, Data Pipelines, Databases, MongoDB, Amazon S3 (AWS S3)

Frameworks

Hadoop, Django, Flask, Next.js

Other

Machine Learning, Data Science, Deep Learning, Computer Vision, System Design, Natural Language Processing (NLP), Computer Vision Algorithms, CSV, Word2Vec, Artificial Intelligence (AI), Data Modeling, Machine Learning Operations (MLOps), Image Processing, Neural Networks, Cloud, Machine Vision, Data Reporting, Data Analytics, Artificial Neural Networks (ANN), Scripting, Deep Neural Networks (DNNs), Software Engineering, Linear Regression, Clustering, Modeling, Data Mining, Back-end, Back-end Development, Large Language Models (LLMs), AI Design, Analytics, Language Models, Data Analysis, CCTV, Architecture, Hugging Face, Fine-tuning, OpenAI GPT-4 API, Workshop Facilitation, Data Scraping, Statistical Methods, Statistical Data Analysis, Statistical Analysis, OpenAI, Llama 2, PEFT, BERT, Software Architecture, Chatbots, Dashboards, Speech to Text, Retrieval-augmented Generation (RAG), Serverless, Model Tuning, ChatGPT Prompts, ChatGPT API, Prompt Engineering, Open-source LLMs, Image Recognition, Convolutional Neural Networks (CNNs), Object Recognition, Llama, Gemini, Gemini API, Edge AI, NVIDIA TensorRT, Quantization, model quantization, NVIDIA Jetson Nano, Text to Speech (TTS), Pattern Recognition, Data Visualization, Statistics, Probability Theory, Numerical Optimization, Optimization, Leadership, Project Leadership, Team Leadership, Fairseq, Transformers, Forecasting, Data Engineering, Automated Data Flows, Cloud Services, Generative Adversarial Networks (GANs), Text Mining, Visualization Tools, Predictive Modeling, Predictive Analytics, Facial Recognition, Big Data, Signal Processing, Reinforcement Learning, Real-time Data, Hardware, Generative Artificial Intelligence (GenAI), Generative Pre-trained Transformers (GPT), Generative Pre-trained Transformer 3 (GPT-3), LoRa, OpenAI GPT-3 API, Financial Modeling, CTO, Object Detection, Object Tracking, Science, Calculus, Programming, Computational Neuroscience, Mathematical Analysis, Robot Operating System (ROS), Microcontrollers, Electronic Data Interchange (EDI), Distributed Systems, Web Scraping, Generative Research, Internet of Things (IoT), eCommerce, Recurrent Neural Networks (RNNs), Amazon Machine Learning, Embedded Systems, DALL-E, Stable Diffusion, Models, Startups, LangChain, CUDA Kernel, 3D Graphics

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.

Share your needs

Discuss your requirements and refine your scope in a call with a Toptal domain expert.

Choose your talent

Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.

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