ML engineer

Pay in the destination countries

Monthly, in US dollars. Each country is an Imora admission package — open one for that route's fees, language requirement and deadline.

What does the day look like?

Your core responsibilities involve designing automated training pipelines, optimizing model inference latency, and deploying ML models as REST and gRPC microservices into production. You work with tools such as Python, C++, PyTorch, TensorFlow, Docker, Kubernetes, Kubeflow, MLflow, Ray, Weights & Biases, and cloud platforms like AWS SageMaker or GCP Vertex AI. Work is structured around a standard 40-hour week, but maintaining real-time production systems (like ad ranking or fraud detection) involves scheduled on-call rotations to handle system outages.

While many imagine the job is about experimenting with cutting-edge architectures and reading research papers, that glamorous side accounts for only 15-20% of your time. The remaining 80-85% is operational engineering: cleaning data, writing unit tests for pipelines, debugging CUDA and GPU driver issues, fixing inference memory leaks, and managing deployment infrastructure.

Will AI replace this job?

ILO 2025 exposure index: 0.53, gradient 3. This is for ISCO-08 2512 Software Developers, on a 0 to 1 task-exposure scale. OECD 2023 also places ISCO major group 2 and sub-major group 25 in the high-exposure, low-risk-of-displacement band. The key point is that these indices measure task exposure, not whether the whole job disappears.

The exposed part is the more routine software work: writing standard unit tests, producing boilerplate training scripts, drafting CI/CD YAML, generating first-pass documentation, and writing baseline data validation queries. That maps directly to the operational half of the day: tests, pipeline work, and data wrangling can be sped up by AI tools. The less exposed part is environment-specific engineering: debugging CUDA driver conflicts and GPU mismatches, tuning inference latency on physical edge hardware, handling live Kubernetes failures, and setting secure on-call escalation rules when a real-time model degrades. Those tasks need system observability, physical constraints reasoning, and operational accountability that generative task models do not run by themselves. In employer demand surveys, WEF 2025 ranks AI and Machine Learning Specialists 1st among the fastest-growing occupations globally and projects positive net employment growth, not job destruction.

What do you need to start?

To enter the field, you need a Bachelor of Science (B.S.) in Computer Science, Applied Mathematics, Software Engineering, or a related quantitative field. For specialized domains such as NLP or Computer Vision, international employers strongly prefer a Master's degree (M.S.), though a PhD is generally reserved for AI Research Scientist roles rather than engineering.

Machine Learning Engineering is an unlicensed and unregulated profession worldwide. Vendor certifications such as AWS Certified Machine Learning - Specialty, Google Cloud Professional Machine Learning Engineer, or Databricks credentials demonstrate practical tool proficiency, but they cannot legally or practically substitute for a university degree when applying for skilled work visa sponsorship.

Where do you study, and for how long?

In Uzbekistan, undergraduate training is offered at Tashkent University of Information Technologies (TUIT) in Artificial Intelligence / Computer Engineering over 4 years (annual basic contract UZS 7,400,000 to UZS 9,000,000 for 2024/2025). Inha University in Tashkent (IUT) teaches Computer Science and Software Engineering in English over 4 years (annual tuition UZS 40,400,000). New Uzbekistan University offers Computer Science / Software Engineering with AI modules in English over 4 years (annual tuition UZS 49,500,000).

Top international graduate destinations include Carnegie Mellon University in the US (MSIIS, 1.5 to 2 years, English, annual tuition USD 54,000 to USD 60,000), ETH Zurich in Switzerland (M.S. in Data Science / Computer Science, 2 years, English, tuition CHF 2,190 per semester / CHF 4,380 per year starting Autumn 2025), Technical University of Munich (TUM) in Germany (M.S. in Informatics / Data Engineering, 2 years, English, EUR 6,000 per semester / EUR 12,000 per year), Imperial College London in the UK (MSc Computing, 1 year, English, overseas tuition GBP 43,250), and KAIST in South Korea (M.S. in AI / Computing, 2 years, English, KRW 4,830,000 per semester, covered by institutional scholarships).

To recognize an Uzbek degree abroad, Germany requires a Statement of Comparability via ZAB (EUR 208 for H+ institutions), while US evaluation through NACES members like WES costs USD 240 to USD 300. Foreign diplomas are recognized in Uzbekistan via my.gov.uz (approximately UZS 680,000). Document apostille is handled by Public Services Centers via e-App.

Where can you actually work?

In Uzbekistan, domestic demand is concentrated within fintech and banking (Uzum Bank, TBC Bank, Kapitalbank), telecoms (Ucell, Beeline), and digital ecosystems (Uzum Market). Local compensation for mid-to-senior engineers ranges from UZS 15,000,000 to UZS 35,000,000 gross per month, though many local roles combine data science and backend tasks rather than pure ML engineering.

Remote work for international companies from Uzbekistan is fully viable via Employer of Record platforms (Deel, Remote.com) or direct B2B contracts, provided you have a proven GitHub track record and deployed production code. Relocation abroad does not operate on government labor quotas; you must secure an individual work visa such as the German EU Blue Card (minimum EUR 41,041 annual salary), the UK Skilled Worker Visa (minimum GBP 38,700), or US H-1B / O-1 visas.

Where are the jobs posted?

Global positions are posted on Levels.fyi Jobs (which filters for visa sponsorship), Y Combinator's Work at a Startup platform (offering access to US/global startups with remote and O-1/H-1B options), the Hugging Face Jobs Board (specializing in NLP and open-source ML), and AI-Jobs.net. You can also apply directly through Big Tech portals: Google Careers, Meta Careers, and Microsoft Careers.

For roles inside Uzbekistan, vacancies are listed on HeadHunter Uzbekistan (hh.uz) and the IT Park portal (it-market.uz). Domestic platforms are completely open to direct applications, whereas international listings often specify whether candidates require existing local work authorization or qualify for visa sponsorship.

How much English, honestly?

University entry requirements vary significantly by country: US graduate programs ask for TOEFL iBT 100 (subscores 25+) or IELTS 7.5; ETH Zurich requires IELTS 7.0 or TOEFL 100; TUM in Germany requires IELTS 6.5 or TOEFL 88; Imperial College London requires IELTS 7.0 (minimum 6.5 per band); and KAIST in South Korea accepts TOEFL 83 or IELTS 6.5.

On the job, you need fluent professional English (CEFR C1) to read technical papers, write documentation, debate architecture trade-offs, and conduct code reviews. While English is entirely sufficient in multinational hubs across Switzerland, Germany, the UK, and the US, local mid-sized European companies often require German (B2), and domestic Asian employers outside international research labs require local fluency (TOPIK 4+ in Korea, JLPT N2/N1 in Japan).

Who hires for this?

Where does this lead?