Available for remote opportunities

Elnur
Shahbalayev

Senior AI & Computer Vision Engineer. I build perception systems that run at the edge — drones, robotics, industrial AI. Currently Team Lead at Bayraktar Teknoloji and AI Engineer at Automata Intelligence.

5+ Years in AI/CV
94% Collision reduction
12M Frames/day processed
47.2 mAP on VisDrone

Engineer by craft,
researcher by instinct

I specialise in real-time computer vision for resource-constrained environments — Jetson-class hardware, drone platforms, industrial edge nodes. My work spans object detection, 3D stereo fusion, anomaly detection, and multi-stream video analytics.

Beyond CV, I'm actively expanding into LLM infrastructure and agentic AI — RAG pipelines, VLM fine-tuning, and production deployment patterns. I believe the next frontier is perception models that reason, not just detect.

I lead an AI perception team at Bayraktar Teknoloji (UAV systems), consult for Automata Intelligence on international government AI engagements, and teach AI/ML at ASOIU. Completing an MSc via a Warwick/ASOIU dual programme.

TensorRT
DeepStream
RF-DETR
YOLOv8/v11
Jetson Orin
ONNX Runtime
PyTorch
GStreamer
RAG Pipelines
LangChain
LLM Fine-tuning
VLMs

Where I've built things

Team Lead — AI Perception
FEB 2024 — PRESENT
Bayraktar Teknoloji · Baku, Azerbaijan
  • Lead perception pipeline for UAV systems: object detection, tracking, stereo depth, collision avoidance
  • Deployed RF-DETR + TensorRT FP16 on Jetson Orin — 94% reduction in collision incidents in field trials
  • Built multi-stream DeepStream pipeline processing 12M+ frames/day across concurrent drone feeds
  • Achieved 47.2 mAP on VisDrone benchmark; integrated frequency-domain enhancement (FEFE module)
  • Designed hierarchical classification loss for military/civilian vehicle distinction
TensorRT DeepStream RF-DETR Jetson Orin Stereo Fusion ONNX
Department Head & AI Engineer
2022 — PRESENT
Automata Intelligence · London (Remote)
  • Lead AI engineering for government and enterprise clients across East Africa and the Gulf
  • Delivered AI coordination layer for eGA (Zanzibar) — pivoted from citizen portal to intelligent API routing system after client discovery
  • Built technical proposals and demo systems for ZRA, ZSSF, ZURA, ADNOC, and MB Group
  • Designed AI Optimizer for ADNOC cogeneration/desalination plant in Ruwais; covered GTG/STG control, KPI methodology, and OT cybersecurity
Agentic AI LLMs Python Dash / FastAPI System Design
AI/ML Instructor
2023 — PRESENT
ASOIU / X Academy · Baku, Azerbaijan
  • Teaching AI/ML Engineering to SABAH Groups (honours cohort) at national technical university
  • Designed full-stack curriculum: Python prerequisites through production model deployment
  • Created original Jupyter notebooks, assessments, and lecture materials from scratch
PyTorch scikit-learn Curriculum Design
AI Engineer (CV / Robotics)
2021 — 2022
Axel Cyber / Risehill · Remote
  • Built real-time warehouse robotics tracking system using instance segmentation and Kalman filtering
  • Developed anomaly detection pipeline for industrial inspection (PatchCore, custom augmentation)
  • Satellite imagery segmentation for agricultural monitoring using U-Net variants
YOLOv8 OpenCV Segmentation PySpark

Things I've shipped

🎯
RF-DETR UAV Detection
Military vehicle detection pipeline for drone footage. TensorRT FP16, custom hierarchical loss, FEFE frequency enhancement. Deployed on Jetson Orin Nano.
47.2 mAP · VisDrone benchmark
RF-DETRTensorRTJetson
🌊
Multi-Stream DeepStream Pipeline
12M+ frames/day multi-camera video analytics on NVIDIA GPU. GStreamer + DeepStream + custom CUDA kernels. Real-time tracking across concurrent feeds.
12M frames/day · 8 concurrent streams
DeepStreamCUDAGStreamer
🤖
eGA AI Coordination Layer
AI routing and orchestration system for Zanzibar e-Government Authority. Intelligent API layer that routes citizen service requests to appropriate government endpoints using LLM classification.
Government-grade deployment
LLMsAgenticFastAPI
🏭
ADNOC Cogeneration AI Optimizer
ML-based optimization system for gas turbine and steam turbine generators at Ruwais desalination plant. KPI-driven control loop with OT cybersecurity constraints.
Industrial OT · GTG/STG optimization
Time-series MLOT SecurityOptimization
📚
Collaborative CV Annotation Tool
Streamlit + Redis + Docker image labeling application for distributed CV annotation workflows. Built to support the Bayraktar dataset pipeline across team members.
Multi-user · Real-time sync
StreamlitRedisDocker
🔬
PV Cell Fault Detection (MSc)
Deep learning pipeline for photovoltaic cell defect detection. Dissertation research at University of Warwick — electroluminescence image classification with uncertainty quantification.
MSc Dissertation · Warwick/ASOIU
PyTorchEL ImagingResearch

Full stack of tools

Computer Vision

  • RF-DETR, YOLOv8/v11, RT-DETR
  • Instance & semantic segmentation
  • Stereo depth & 3D fusion
  • Object tracking (ByteTrack, DeepSORT)
  • Anomaly detection (PatchCore)

Edge Deployment

  • TensorRT (FP16/INT8)
  • ONNX Runtime
  • NVIDIA DeepStream
  • Jetson Orin / Nano
  • GStreamer pipelines

LLM & Agentic AI

  • RAG pipeline design
  • LangChain / LlamaIndex
  • VLM fine-tuning (LLaVA, Qwen-VL)
  • Prompt engineering
  • Agent orchestration

ML Engineering

  • PyTorch (custom losses, training loops)
  • Hugging Face ecosystem
  • MLflow, Weights & Biases
  • Docker, CI/CD
  • FastAPI, Streamlit

Let's build
something

Open to senior/staff remote roles in computer vision, ML engineering, and LLM infrastructure. UTC+4, available for EU/UK hours.

elnur@shahbalayev.com