Why Data, ML and AI Projects Are Completely Useless

Summary of the video’s main arguments

• Projects are largely ineffective for getting a job in data/ML (unless they’re full SaaS-level products). The presenter argues that for most people—especially experienced professionals—portfolio projects do not meaningfully trigger hiring decisions. Employers value proven experience far more.

• If you already have real experience (e.g., at least one internship), projects won’t help you get hired. The core claim is that what matters most is demonstrating you can do the job already—something internships and professional track records do better than side projects.

• The real gap isn’t technical skill—it’s positioning and job-search execution. The presenter says universities teach technical work (building thesis-style projects), but not how to get interviews and offers. Instead, job success depends on:

  • resume and LinkedIn positioning focused on results and impact
  • effective networking
  • finding hidden opportunities
  • interview techniques that convert interviews into offers

• A “projects-only” mindset is framed as self-deception. He dismisses the idea that recruiters will automatically notice your GitHub project and then offer an interview (e.g., after applying and being rejected, then publishing a project and expecting an immediate callback).

• Timing/luck and market changes matter; what worked before may not work now. He notes that in earlier periods projects were more helpful, but the market has become stricter and more oversaturated. Therefore, strategies relying on projects have diminishing returns.

• For people without experience, most projects still won’t get hired—unless they create real business/financial impact. He says the only projects that might help are unusually impactful ones (e.g., building a SaaS product, or algorithms that affect decisions or money). But he argues these take a lot of time and effort.

Recommended strategy for beginners

• Freelance as “productive practice with proof.” Instead of Kaggle-style competitions, he suggests using freelance platforms (Fiverr, Upwork, PeoplePerHour) to do data/ML work for a client, initially often for free. The goal is to gain real-world experience that can be credibly described on a resume/LinkedIn.

Courses/projects are not denied—just separated by purpose

• Learning is useful, but job acquisition depends on job-search mechanics and evidence of impact. He supports using courses and projects to learn or for fun, but argues they won’t be the main driver of employment outcomes.

Presenters / contributors

• Egor (the main presenter/author of the advice; referenced as “Egor” throughout)