NazrinAtayeva

Machine learning engineer for computer vision, 3D perception and LLM agents.

Before my MSc in AI for Engineering at the University of Sheffield in the United Kingdom, I spent almost three years shipping iOS and React Native apps for banking and telecom. I still work like a product engineer: tests, tracing, and models that run outside a notebook.

Based in Baku, UTC+4. Open to remote roles across Europe.

In range
0
Detected
0
Confidence
0.00
  • Car
  • Truck
  • Pedestrian
  • Bicycle
  • Barrier
  • Cone
  • Teacher's box
  • Missed
Teacher: BEVFusion, the full-size camera and LiDAR model. Boxes are tight, steady and confident. A simulated 32-beam LiDAR scan, showing the detection task from my MSc dissertation. Rings are 10 m apart.

Selected work

Four projects across 3D perception, LLM agents and time-series machine learning. The code lives on GitHub.

MSc dissertation, University of Sheffield, United Kingdom, 2026

Knowledge distillation for BEVFusion

BEVFusion merges camera images and LiDAR into one bird's-eye-view map to find cars, people and obstacles in 3D. It is accurate and heavy. I distilled it into a compact student built from YOLO and PointPillars, and benchmarked it on nuScenes.

  1. Student on its own
  2. Class weights for rare objects
  3. Learning from the teacher
  4. Both combined
  • PyTorch
  • nuScenes
  • Docker
  • CUDA
  • RTX 3090

One agent turn, traced with OpenTelemetry

  1. agent.turn
  2. compare_pace
  3. get_tyre_degradation
  4. simulate_pit_window
  5. llm.generate
  6. guardrail.check

Span order is real; timings are illustrative.

Agentic LLM system

Race Engineer Agent

An LLM that plays Formula 1 race engineer. It reads FastF1 timing data, calls tools to compare pace, model tyre wear and simulate pit windows, then decides what to tell the driver, including when to say nothing.

  • Runs on Gemini, or locally on Qwen 3 14B
  • Guardrails against repeated messages and bad pit calls
  • Instrumented end to end and covered by tests
  • Python
  • FastF1
  • Ollama
  • OpenTelemetry

Three detectors on the same sensor channel

Illustration of the comparison view.

Time-series anomaly detection

Sensor anomaly detection

Flags faults in Formula 1 sensor telemetry. Three detectors run side by side, so you can see where they agree and where they don't.

Open live demo
  • Python
  • Isolation Forest
  • One-Class SVM
  • LSTM
  • Streamlit

Features for a race come only from races before it

  • Used for features
  • Race being predicted
  • Not yet run

Tabular classification

Points-finish predictor

Predicts whether a driver finishes in the points, using only what was known before each race. I also tracked how the model holds up as the sport changes over time.

Open live demo
  • Python
  • Feature engineering
  • Drift analysis
  • Streamlit

Experience

I moved into machine learning after almost three years as a mobile engineer. Production apps taught me to test, measure and ship.

  1. Until 2025

    Unibank, Azerconnect, Morpho

    Mobile engineer

    iOS apps in UIKit and SwiftUI, and cross-platform apps in React Native, for banking and telecom products used in production.

  2. 2025–2026

    University of Sheffield, United Kingdom

    MSc Artificial Intelligence for Engineering

    Distinction, 74% overall. Dissertation on knowledge distillation for camera and LiDAR 3D object detection. Degree conferred January 2027.

  3. Now

    Baku, working remotely

    Looking for my next role

    Machine learning, computer vision and AI engineering, remote across Europe.

Toolkit

The frameworks I build with, grouped by the kind of problem they solve.

Perception

Finding and locating objects in 3D from camera images and LiDAR.

Frameworks

  • PyTorch
  • CUDA
  • YOLO
  • PointPillars
  • nuScenes

Skills

  • 3D object detection
  • Camera and LiDAR fusion
  • Knowledge distillation

Used in my MSc dissertation

Machine learning

Models for tabular and time-series data, from features to evaluation.

Frameworks

  • scikit-learn
  • pandas
  • NumPy
  • MATLAB

Skills

  • Anomaly detection
  • Time-series analysis
  • Feature engineering

Used in anomaly detection and the points predictor

LLM systems

Agents that call tools, stay on task and can be traced.

Frameworks

  • Gemini API
  • Ollama
  • OpenTelemetry

Skills

  • Tool-calling agents
  • Guardrails
  • Local models

Used in Race Engineer Agent

Engineering

Getting models out of notebooks and into tested, shippable code.

Frameworks

  • Python
  • Docker
  • Git
  • pytest
  • Streamlit

Skills

  • Automated testing
  • Cloud GPU training
  • SQL

Used in every project

Mobile

Almost three years of production iOS and React Native apps.

Frameworks

  • Swift
  • SwiftUI
  • UIKit
  • React Native

Skills

  • Banking and telecom apps
  • Shipping to production

Used at Unibank, Azerconnect and Morpho

Contact

Hiring for machine learning, computer vision or AI engineering? Email is the fastest way to reach me.

Location
Baku, Azerbaijan
Local time
--:-- in Baku
Elsewhere
GitHubLinkedIn