What Do Machine Learning Engineers Do?

1 week ago

ARTICLE AD BOX

In this article, I want to explicate what I do arsenic a instrumentality learning engineer.

The purpose is to thief anyone looking to participate nan section summation a truthful position of what a instrumentality learning technologist is, really we work, what we do, and what a emblematic time successful life is like.

I dream it tin thief you pinpoint if a profession successful instrumentality learning is so for you.

What is simply a instrumentality learning engineer?

Due to nan accelerated acceleration of nan tech/AI space, a instrumentality learning technologist is still not well-defined and varies betwixt companies and geographies to a definite extent.

However, it mostly refers to personification who:

Mixes instrumentality learning, statistic and package engineering skills to train and deploy models into production.

At immoderate companies, location will beryllium a ample cross-over pinch information scientists. Still, nan main favoritism betwixt nan 2 roles is that instrumentality learning engineers present nan solution into production. Often, information scientists won’t do this and attraction much connected helping successful nan model-building stage.

The request for a instrumentality learning technologist came from nan truth that models successful Jupyter Notebooks person zero value. So, a domiciled well-versed successful instrumentality learning and package engineering was needed to thief bring nan models “to life” and guarantee they make business value.

Because of this wide skillset, instrumentality learning engineering is not an entry-level role, and you would typically request to beryllium a information intelligence aliases package technologist for a mates of years first.

So, to summarise:

Responsibilities: Train, build and deploy instrumentality learning models.
Skills & Tech: Python, SQL, AWS, Bash/Zsh, PyTorch, Docker, Kubernetes, MLOps, Git, distributed computing (not an exhaustive list).
Experience: A mates of years arsenic a information intelligence aliases package engineer, and past up-skill yourself successful nan different areas.

If you want a amended knowing of nan different information and instrumentality learning roles, I urge checking retired immoderate of my erstwhile articles.

The Difference Between ML Engineers and Data Scientists
Helping you determine whether you want to beryllium a information intelligence aliases instrumentality learning engineermedium.com

Should You Become A Data Scientist, Data Analyst Or Data Engineer?
Explaining nan differences and requirements betwixt nan various information rolesmedium.com

What do I do?

I activity arsenic a instrumentality learning technologist wrong a cross-functional team. My squad specialises successful classical instrumentality learning and combinatorial optimisation-based problems.

Much of my activity revolves astir improving our instrumentality learning models and optimisation solutions to amended nan customer acquisition and make financial worth for nan business.

The wide workflow for astir of my projects is arsenic follows:

Idea — Someone whitethorn person an thought aliases presumption astir really to amended 1 of our models.
Data — We cheque if nan information to beryllium aliases disprove this presumption is readily disposable truthful we tin commencement nan research.
Research — If nan information is available, we commencement building aliases testing this caller presumption successful nan model.
Analysis — The results of nan investigation shape are analysed to find if we person improved nan model.
Ship — The betterment is “productionised” successful nan codebase and goes live.

Along this process, location is simply a batch of relationship pinch different functions and roles wrong nan squad and broader company.

The thought shape is simply a collaborative chat pinch a merchandise head who tin supply business penetration and immoderate captious impacts we whitethorn person missed successful nan first scoping.
Data, Build, and Analysis tin beryllium done successful collaboration pinch information analysts and engineers to guarantee nan value of our ETL pipelines and nan usage of nan correct information sources.
The investigation conception would usage nan thief of information scientists to usage statistic and instrumentality learning skills erstwhile looking to amended our model.
The vessel shape is simply a associated effort pinch our dedicated package engineers, ensuring our deployment is robust and up to modular pinch champion coding practices.

From experience, I cognize that this type of workflow is prevalent among instrumentality learning engineers successful galore companies, though I americium judge location are flimsy variations depending connected wherever you are.

My occupation is besides not conscionable to constitute codification time successful and time out. I person different responsibilities, for illustration conducting workshops, presenting to stakeholders, and mentoring much inferior members.

What is nan building of instrumentality learning teams?

Machine learning engineers activity successful galore different ways crossed an organisation, but location are 3 chopped options, and nan remainder are a operation of them.

Embedded — In this case, instrumentality learning engineers are embedded successful cross-functional teams pinch analysts, merchandise managers, package engineers and information scientists, wherever nan squad solves problems successful 1 domain wrong nan company. This is really I work, and I really for illustration it because you get to prime up tons of valuable skills and abilities from different squad members who are specialists successful their ain right.
Consultancy — This is nan flip side, wherever instrumentality learning engineers are portion of an “in-house consultancy” and are portion of their ain team. In this scenario, nan instrumentality learning engineers activity connected problems based connected their perceived worth to nan business. You are technically little specialised successful this action arsenic you whitethorn request to alteration nan type of problems you activity on.
Infrastructure/Platform — Instead of solving business problems directly, these instrumentality learning engineers create in-house devices and a deployment level to make productionising nan algorithms overmuch easier.

All ways of moving person pros and cons, and successful reality, I wouldn’t opportunity 1 is amended than nan other; it’s really a matter of individual preference. You still do breathtaking work, nonetheless!

What is simply a emblematic time successful a life?

People online often glamourise moving successful tech, for illustration it’s each java breaks, chats, and coding for an hr a day, and you make good complete six figures.

This is decidedly not nan case, and I wish it was true, but it’s still a nosy and enjoyable workday compared to galore different professions.

My wide acquisition has been:

9:00 am — 9:30 am. Start astatine 9 americium pinch a greeting standup to drawback up pinch nan squad regarding nan erstwhile day’s activity and what you are doing today. A “standup” gathering is very communal crossed tech.
9:30 am — 10:30 am. After nan standup, location whitethorn beryllium different gathering for an hour, 9:30–10:30 aliases so, pinch stakeholders, engineers, an all-hands aliases different institution meetings.
10:30 am — 13:00 pm. Then, it’s a work/code artifact for 2 hours aliases truthful wherever I attraction connected my projects. Depending connected my work, I whitethorn brace pinch different information scientist, instrumentality learning technologist aliases package engineer.
13:00 pm — 14:00 pm. Lunch.
14:00 pm — 17:45 pm. Afternoons are usually free of meetings, and location is simply a ample artifact of attraction clip to activity connected your projects. This is chiefly for individual contributors for illustration myself.
17:45 pm — 18:00 pm. Reply to emails and Slack messages and wrap up for nan day.

Every time is different, but this is what you tin expect. As you tin tell, it’s thing “extrordinary.”

This is besides nan workday for a inferior / mid-level individual contributor (IC) for illustration myself. Senior positions, particularly managerial roles, typically person much meetings.

An important point to statement is that I don’t ever codification successful my activity blocks. I whitethorn person a position to hole for stakeholders, immoderate ad-hoc study for our merchandise manager, aliases immoderate penning up of my latest research. I whitethorn not moreover codification for nan full day!

On average, I walk 3–4 hours difficult coding; nan remainder is meetings aliases ad-hoc work. Of course, this varies betwixt companies and astatine different times of nan year.

Why americium I’m a instrumentality learning engineer?

The logic I americium a instrumentality learning technologist tin beryllium boiled down to 4 main reasons:

Interesting. As a instrumentality learning engineer, I get to beryllium correct astatine nan forefront of nan latest tech trends for illustration AI, LLMs, and beautiful overmuch thing that is going viral successful nan field. There is ever thing caller and breathtaking to learn, which I love! So, if you want to perpetually study caller skills and use them, this whitethorn beryllium a profession you would beryllium willing in.
Work-Life Balance. Tech jobs mostly supply amended work-life equilibrium than different professions for illustration banking, rule aliases consulting. Most instrumentality learning jobs are 9–6, and you tin often walk a fewer days moving from home. This elasticity allows maine to prosecute different passions, projects, and hobbies extracurricular of work, specified arsenic this blog!
Compensation. It’s nary concealed that tech jobs supply immoderate of nan highest salaries. According to levelsfyi, nan median net of a instrumentality learning technologist successful nan UK is £93k, which is crazy for an mean value.
Range of Industries. As a instrumentality learning engineer, you tin activity successful loads of different industries during your career. However, to go a existent specialist, you must find and instrumentality to 1 manufacture you love.

I dream this article gave you much penetration into instrumentality learning, if you person immoderate questions fto maine cognize successful nan comments.

Another thing!

Join my free newsletter, Dishing nan Data, wherever I stock play tips, insights, and proposal from my acquisition arsenic a practicing information scientist. Plus, arsenic a subscriber, you’ll get my FREE Data Science Resume Template!

Dishing The Data | Egor Howell | Substack
Advice and learnings connected information science, tech and entrepreneurship. Click to publication Dishing The Data, by Egor Howell, a…newsletter.egorhowell.com