◆ Data & Analytics

What a machine learning engineer
really does.

23 tasks, each one witnessed by the sources that watched the job — and behind every one, a prompt you can use tonight.

23evidenced tasks
262,440in the US (2025)
$120,230median pay / year
10systems it runs on
See the 23 tasks ↓Open in the atlasThe full career map →
This is what one task looks like here
Manage large amounts of data
Provision a fault-tolerant storage and processing pipeline for terabyt…3 sources agree

The shape of the day

tap a movement to see its tasks

Which one is you, right now?

Pick the moment · no score, no sign-up
Which moment is you right now?
Whichever you pick, the task behind it opens below.

The work, task by task

23 tasks
Hands on the work19
Manage large amounts of data+
Provision a fault-tolerant storage and processing pipeline for terabytes of sensor and user event data, define retention and partitioning strategy, and set alerts for backpressure and job failures before the monthly release.
escoonetwiki3 agree
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Present findings through reports and presentations+
Prepare a ten-slide deck for Friday’s review showing experiment setup, key metrics, segment lift, failure examples, and a one-paragraph recommendation for next steps, and circulate to product and analytics with a read request.
escojdonet3 agree
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Communicate insights to stakeholders+
Summarise the model findings, risks, and business implications for the recommender project in a two-page briefing for Priya in product and Jamal the head of ops, include key charts, one-slide ask, and a one-week decision deadline.
escojdonet3 agree
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Apply machine learning techniques+
Train and validate the new collaborative filtering pipeline on the user-item dataset, log experiments, compare precision@10 and recall@10 to last release, and hand over the best model and training notes to the engineering lead by Wednesday.
jdonet2 agree
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Merge data sources+
Ingest user logs, product catalog, and sales transactions, deduplicate and standardise identifiers, join into a single analytics table keyed by user_id and product_sku, and produce a daily refreshed dataset for the recommender team by Friday.
escojd2 agree
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Determine available and useful data for projects+
Inventory internal and public datasets we can access for the next recommender proof of concept, note schema fields, freshness, permissions, and sample quality, then recommend which three sources to onboard first with estimated effort.
jd
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Clean and process raw data+
Take the raw user logs, product metadata, and interaction CSVs, remove duplicates, normalise timestamps, impute missing identifiers, and output a clean feature table ready for model training with row counts and error logs.
jd
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Interpret data analysis results+
Review the latest model run, compare predicted versus actual engagement by cohort and item, calculate precision, recall, calibration and AUC, and summarise anomalies and actionable next steps for the product owner.
jd
when the reply comes backPush once: ask it to sharpen the weakest part, and to say what it assumed. Helpful?
Watch and assess3
Grow the practice1

What the work runs on

named inside the evidenced tasks
12 tasksApache Sparkfast iterative training and validation across large cleaned samples
4 tasksApache Kafkaingest and buffer high-throughput event streams into the pipeline
4 tasksAtlassian Confluenceshareable narrative and inline visuals for cross-team review and comments
3 tasksAtlassian JIRAtrack follow-up actions and assign decisions to stakeholders
3 tasksApache Hivequeries and profiles table schemas and freshness across data lakes to assess source usefulness
1 taskApache Hadoopscalable distributed storage and batch processing for very large datasets
1 taskApache Cassandraproduces consistent, queryable key-value views for normalized user and product attributes
1 taskAmazon Elastic Compute Cloud EC2runs the statistical calculations and extraction scripts at scale against production datasets
Browse every evidenced software task →

The same task, four heights

this page is height one
ExecuteDo today's task, with fewer mistakesyou are here → ImproveMake it easy for the next person to acceptin the atlas → DecideWork out the right move when it is unclearin the atlas → BecomeLearn the pattern so it stops coming backin the atlas →

Can AI actually do this job?

the honest answer

It can

where it genuinely helps
  • Explain the theory behind the work
  • Draft, tidy and structure your writing
  • Rehearse a hard conversation before you have it
  • Build a study plan that fits your gaps

It cannot

where it stops, completely
  • Be in the room where a machine learning engineer actually works
  • Carry the responsibility when the call is wrong — that weight stays yours
  • Notice what no one wrote down: the hesitation, the thing left unsaid
  • Live with the outcome

What the work pays

two countries, two different measures

United States

this exact occupation · BLS 2025
  • $120,230 a year — the middle: half earn more, half earn less
  • The lowest tenth earn near $67,240; the top tenth near $199,130
  • 262,440 people employed in this occupation

India

the occupation GROUP, not this job · PLFS via ILOSTAT 2025
  • ₹38,298 a month — the median for Professionals, the group this work sits in
  • India publishes pay by broad occupation group, so this covers many jobs besides this one. It is a shape, not a salary.
read this carefullyThese two numbers are not comparable and must not be converted into each other. One is a yearly figure for this job alone; the other is a monthly figure for a whole family of jobs. What travels between them is the pattern, not the amount: experience lifts pay almost everywhere.

Where the evidence lives

open any of it yourself
O*NETThe US Department of Labor's own record of this job9 of these tasks ESCOThe European skills and occupations register9 of these tasks Job descriptionsReal postings, in the wild14 of these tasks WikipediaHow the world describes this work5 of these tasks US Bureau of Labor StatisticsPay and employment, by occupation2025 figures UK NOS TECIS805301Assist in developing and validating machine learning solutionsnational standard

Close to this work

12 nearby
Data & AnalyticsFinancial Analyst26 evidenced tasks Data & AnalyticsExperimentation Analyst26 evidenced tasks Data & AnalyticsData Governance Analyst26 evidenced tasks Data & AnalyticsR Analyst26 evidenced tasks Data & AnalyticsCrm Analyst26 evidenced tasks Data & AnalyticsWeb Analyst26 evidenced tasks Data & AnalyticsGrowth Analyst26 evidenced tasks Data & AnalyticsProduct Analyst25 evidenced tasks

Questions people actually ask

You’ll split time between coding models and working with data. Mornings often mean checking model jobs on Apache Airflow, looking at logs in Amazon EC2 instances, or restarting Spark jobs if a pipeline failed.

Afternoons usually go to meetings: syncs in JIRA or Confluence with product and data teams, reviewing model metrics, and planning next experiments. Evenings may be for reading new papers or tuning models and retraining on cleaned data.

Start with Python plus Apache Spark for big-data processing and SQL/Hive for querying large datasets (Apache Hive). Learn how to run jobs on Amazon EC2 and schedule workflows with Apache Airflow.

Add Kafka basics for streaming data, and a ticketing/wiki tool like Atlassian JIRA and Confluence so you can communicate work and document models.

The U.S. Bureau of Labor Statistics reports about 262,440 employed machine learning engineers with a median pay of $120,230 per year.

They list the lowest tenth at $67,240 and the top tenth at $199,130. Those numbers are from BLS national statistics and cover many industries and locations.

Build checks: test model fairness and performance across groups, log inputs and outputs, and run monitoring pipelines (Airflow + Spark) so you can catch drift or bias quickly.

Use Confluence or JIRA to record data sources (Apache Cassandra, Hive), assumptions, and approval steps. If a model affects people, involve legal or compliance early and keep reproducible code and datasets.

Practical experience building and keeping models running in production matters most: know how to clean data, train models, and monitor them using Spark, Airflow, Kafka, and EC2.

Hiring managers look for examples where you improved model performance, fixed data problems, or scaled a pipeline — not just theory. Show code, logs, and before/after metrics.

Practice the full pipeline: merge sources, clean data, and build models locally with Python and small Spark or Pandas jobs. Use SQLite or a local Hive-like setup to practice queries.

Simulate production by scheduling retrains with Airflow, stream small events through Kafka locally, and deploy a model to a cheap EC2 instance to learn monitoring and scaling.