Beyond the Code: How Abylaikhan Azamatov Is Building Human-Centred AI Through Hackathons, Social Apps, and Digital Empowerment

Optimizing deep learning models for constrained environments is one of the most pressing challenges in applied machine learning. According to a comprehensive survey of edge AI research, over 70% of deployment failures occur due to mismatches between model complexity and limited hardware capabilities on edge devices. Many industries require computer vision and automation solutions that can function on low-power devices, under unstable connectivity, and with real-time requirements. While cutting-edge AI often relies on powerful hardware and ideal conditions, real-world deployment demands something different: smart engineering that balances performance, speed, and scale.

But while many organizations struggle with this gap, engineers like Abylaikhan Azamatov are already creating solutions that prioritize real users. A 25-year-old AI specialist from Kazakhstan, Azamatov has completed over a dozen machine learning projects across fintech, industrial services, and consumer wellness. A graduate of the International Information Technology University with certifications in computer vision and deep learning from Stanford and DeepLearning.AI, he specializes in compressing models, optimizing inference pipelines, and developing systems capable of stable 10 FPS video stream analysis and up to 95% classification accuracy under non-ideal conditions.

His portfolio includes a DeepFake detection model developed during the Kryptonite ML Challenge, where he engineered a multi-stage architecture combining CNN-based image feature extraction with recurrent layers for temporal anomaly detection. For KazVysotStroy, he designed an automated task allocation system for industrial climbers, using priority queues and asynchronous resource matching algorithms to replace manual scheduling. The system reduced task dispatch time from hours to seconds and eliminated 100% of in-person document signing via digital contract verification tools.

Another key project is LadyGym App—a mobile fitness platform for women's gyms in the CIS. While business-facing on the surface, the backend is highly technical: QR-based user validation, GPS-based training verification, and a gamified reward engine supported by real-time user segmentation and engagement tracking. Built using Flutter for mobile and Node.js/PostgreSQL for backend, the platform automated 100% of previously manual gym operations and achieved 30% fewer client complaints due to scheduling errors in the first two months post-launch.

Having worked across Kazakhstan, Russia, Japan, and the UAE, Azamatov adapted his models to various infrastructure realities—from low-bandwidth edge computing setups using NVIDIA Jetson Nano to cloud-integrated GPU pipelines with batch processing for finance applications. His journey from winning the LATOKEN Hackathon (with a predictive blockchain transaction validator) to joining the Hackathon Raptors Association illustrates how engineering precision combined with real-world pragmatism can bring AI out of the lab and into people's lives.

Computer Vision Applications in Industrial Settings

Azamatov's early participation in hackathons laid the groundwork for his practical approach to AI. By participating in events like HackDay at the International IT University in Almaty, the Kryptonite ML Challenge on DeepFake detection, and the Siam ML Hack focused on petroleum data, he learned to combine theory with practical applications. However, the international LATOKEN Hackathon marked a turning point: his team not only won the challenge but also earned him a job offer from the organizing fintech company.

"Hackathons train your ability to validate hypotheses under time constraints," Azamatov explains. "When developing DeepFake detection models, for example, you don't have time to train massive networks from scratch—you prioritize pre-trained backbones, freeze early layers, and focus on tuning classification heads. It's all about optimizing the pipeline for performance under limited computational resources." He later joined Hackathon Raptors Association, a professional AI and data science community, to stay connected with top minds and remain at the forefront of practical ML development.

Digital Tools for Real-World Labor: The KazVysotStroy Project

When KazVysotStroy—a leading company in industrial climbing services in Central Asia—faced operational bottlenecks, Azamatov offered a digital answer. He led the development of an integrated platform that connects project managers with professional climbers, replacing manual call-outs and in-person paperwork with automated workflows.

The solution includes a mobile app for climbers, a custom CRM for managers, and end-to-end digital document management. Work orders are distributed using asynchronous resource-matching algorithms based on real-time availability and task priority. Document exchange and payment confirmation are fully digitized using QR verification and time-stamped records.

The optimization of scheduling algorithms reduced average task assignment time from 2–3 hours to under 45 seconds. Backend processing latency decreased by 67%, and document signing time dropped to near-zero due to integrated digital contract modules. This level of automation also enabled real-time revenue tracking and digital audit trails, turning an analog workflow into a scalable, auditable system.

LadyGym App: A Fitness Ecosystem Designed for Women

Azamatov's work extends beyond industrial sectors. LadyGym, a fitness application developed exclusively for women's gyms in the CIS region, stands as a testament to its commitment to inclusive technology. The idea was born from the everyday chaos of managing fellowships, staff, and class attendance via spreadsheets and paperwork.

LadyGym digitized these processes through a modular platform: QR-based check-ins for clients and trainers, geolocation-verified training sessions, gamified progress tracking, and a CRM for fellowship and payment management. For QR recognition, the mobile app used OpenCV in combination with the ZXing (Zebra Crossing) barcode scanning library, enabling fast decoding even under low-light conditions or partial occlusion. Geolocation validation was handled through GPS polling combined with boundary logic to prevent spoofing.

Built with Flutter for the mobile interface and Node.js + PostgreSQL for backend infrastructure, the system also featured real-time data syncing and push notifications. The app saw over 300 installs in its first few months.

"Implementing geolocation consistency checks without degrading client-side performance was a key challenge," Azamatov explains. "We had to optimize background location polling intervals and compress updates using protobuf messaging to keep mobile resource usage below 5%."

Certified to Lead: Education and Expertise

Azamatov's practical success is rooted in solid academic rigor. He earned computer vision and deep learning certifications through Coursera, including from Stanford and DeepLearning.AI programs. A specialized international course in computer vision added further depth to his skill set, blending theoretical expertise with production-ready implementation.

"Those certifications didn't just teach me technology—they taught me structure. And that's what separates hobby projects from scalable systems," he reflects.

From Projects to Principles: Building Ethical, Purposeful AI

In all his work, Azamatov champions an approach where technology serves the user, not vice versa. His designs prioritize usability, autonomy, and respect for real-world constraints. Whether digitizing workflows in harsh industrial settings or enabling fitness journeys in women-only gyms, he frames AI as a tool for clarity, not complexity.

As he continues to explore new areas, Azamatov sees AI not just as a profession, but as a responsibility. "When building for low-resource environments, you constantly make trade-offs between model complexity and execution time, accuracy and interpretability, flexibility and latency," he says. "My priority is always controlled performance: achieving maximum utility with minimum overhead."

In a region where digital transformation is still unfolding, Azamatov's work proves that AI doesn't have to be massive or global to be meaningful. It just needs to work for real people, in real places, solving real problems. And that's precisely what he's doing.