08/04/2026 updated
IP
100 % available
Senior Machine Learning Engineer
Sofia, Bulgaria
Worldwide
Management Information Systems, Mannheim, M.Sc.Programming, Python, C/C++, C#, Java, Lua, SQL, Deep learning, computer vision, reinforcement learning, NLP, analytics, Tensorflow, Jupyter, pandas, NumPy, Keras, scikit-learn, Spark, Beam, Cloud, GCP, Kubernetes Engine, Compute Engine, Storage, BigQuery, StackDriver, BigTable, AWS, EC2, S3, ECS, Containers, orchestration, Kubernetes, Docker, MySQL, Redis, Microsoft SQL, Scrum, Kanban, distributed computing, Blockchain, data processing, Machine Learning, Robotic control, anomaly detection, Cloud GPU/TPU, GCP BigQuery, Google, algorithms, Google Cloud infrastructure, scalability, bootstrapping, algorithmic trading, image recognition, low-latency
Experience
Professional experience10+ years
Languages
BulgarianFluentGermanFluentEnglishFluent
Project history
Delivered machine learning projects from problem discovery to production for the following applications:
● Robotic control system using learning from human demonstration
● Camera-based vehicle damage detection and inspection of warehouse containers
using deep learning
● Computer vision based product verification and detection
● Natural language models for chat support
● Bot failure prediction and anomaly detection using time series models
Developed cloud infrastructure for large-scale data processing, model training and lifecycle management (Kubernetes, Cloud GPU/TPU, GCP BigQuery, Storage, Tensorflow, Beam)
Two patent filings, consultancy of other teams in numerous robotics, computer vision and time series projects.
Google DeepMind, UK
Developed a platform for distributed training of reinforcement learning and deep learning
algorithms in Lua/ Torch running on 10K+ machines on Google Cloud infrastructure used by
several groups in the company.
Designed algorithmic changes for improving the data-efficiency and scalability of reinforcement
learning algorithms and achieving state-of-the-art results in dexterous manipulation
benchmarks.
First-author paper submission and a patent filing, "Data-efficient Deep Reinforcement Learning
for Dexterous Manipulation", Popov et al., 2017.
Created techniques for using demonstration data, reward shaping and starting state
exploration bootstrapping.
Developed a platform for distributed training of reinforcement learning and deep learning
algorithms in Lua/ Torch running on 10K+ machines on Google Cloud infrastructure used by
several groups in the company.
Designed algorithmic changes for improving the data-efficiency and scalability of reinforcement
learning algorithms and achieving state-of-the-art results in dexterous manipulation
benchmarks.
First-author paper submission and a patent filing, "Data-efficient Deep Reinforcement Learning
for Dexterous Manipulation", Popov et al., 2017.
Created techniques for using demonstration data, reward shaping and starting state
exploration bootstrapping.
Point Ltd and other (contractor)
Developed deep learning models for time series prediction, algorithmic trading, natural
language processing and image recognition.
Developed pricing and risk models for an innovative real estate equity product for a US
FinTech start-up.
Developed deep learning models for time series prediction, algorithmic trading, natural
language processing and image recognition.
Developed pricing and risk models for an innovative real estate equity product for a US
FinTech start-up.