Author: Shubajit Saha

Disclaimer: All opinions expressed in this document are my own.

Untitled

                                                                   Source: [xkcd](<https://www.google.com/url?q=https://xkcd.com/2173/&sa=D&source=editors&ust=1703276365054508&usg=AOvVaw1z7Atpr0z-HfuP5mWiZbWp>)

What are the job responsibilities of an ML Engineer?

ML Engineering roles are relatively new in the industry and hence the exact job responsibilities depends on the company, team etc. Some companies treat ML skillset differently than SWE skills (having separate teams for ML research and productionization) where as in some companies ML is just considered a specialization area for SWEs.  (Teams are more interleaved and no clear segregation of research and productionization groups)

The ideal MLE role contains a mix of both SWE tasks and building ML models. On one side of the spectrum we will have a role which requires you building ML infra (involving mostly coding), on the other hand we will have a role that just focus on analyzing datasets and building models (more research oriented, very less coding). based on your skills and team requirements, you decide what percentage of time you spend on coding(writing code for infra) vs research (building models/analysis/new techniques)

Some of the tasks that I do (have done) as part of my job:

  1. Understand a business problem and convert that to a ML problem (figuring out the training data collection, label collection strategy, estimating the acceptable performance of the model etc.)
  2. Experiment with new features for existing models and analyze model performance.
  3. Productionizing the signals by writing necessary code to plug them into existing stack.
  4. Train models withe new dataset and analyze model performance.
  5. Deploy approved models to production [mostly involves config changes].
  6. Writing batch jobs for feature generation, training data processing, metric computation.
  7. Doing open ended analysis to understand current model gaps and potential area of improvement.
  8. Monitor performance and health of existing ML infra and models in productions.
  9. Work on escalations by product stakeholders.

What are the skills required for MLE roles?