ML Engineer · Step 8 of 8 · 6 min
Responsible ML and your portfolio project
You can check a model for bias and privacy risks, and you have a plan for one project that shows you can do the job.
Bias comes from data and choices
A model learns patterns in its data, including unfair ones. A hiring model trained on past hires can learn to prefer the groups that were hired before. Removing the 'gender' column is not enough, because other features, such as a postcode or a hobby, can stand in for it.
The practical habit: measure performance separately for each group that matters, such as age bands, regions or languages. If error rates differ a lot, find out why before you ship. Document what the model is for and what it should not be used for, in a short model card.
Privacy and the law
Collect only the data you need, remove names and IDs where you can, and control who can see raw data. Do not paste customer data into public chatbots. Rules like Europe's GDPR and the EU AI Act, and Vietnam's personal data protection rules, set duties for how data is used and how high-risk systems are checked.
You do not need to be a lawyer. You do need to ask early: where did this data come from, did people agree to this use, and who could be harmed if the model is wrong?
One project, done properly
Employers want proof you can take a problem all the way. One complete project beats ten copied tutorials. Pick a real question with public data, for example from Kaggle, a government open-data portal or Hugging Face datasets.
Show the full path: a clear question, a baseline, honest evaluation with the right metric, a model behind a small API or demo app (Gradio and Streamlit are free and quick), and a short section on limits and risks. Put the code on GitHub with a README a stranger can follow.
Make it easy to judge
Start the README with three lines: the problem, the result, and a link to try it. Hugging Face Spaces hosts small demos for free. Then write a short blog post on what went wrong and what you learned. Reviewers remember honesty about failure more than a high score.
Try it
Review your project plan
- 1.Take the project plan you wrote in the workflow step, or write a new one in a few lines.
- 2.Paste it into a chatbot with the prompt below.
- 3.Fix the two weakest points it finds before you start building.
Act as a hiring manager for ML engineers. Review this portfolio project plan: [paste plan]. Point out risks of data leakage, bias or privacy problems, check that the metric fits the problem, and say what would make the project stand out. Be direct and list the two most important fixes.