Breaking into AI research from a software engineering background is not for the faint of heart, and a new blog post from a former Meta engineer proves it with receipts. The author, who spent three years as a senior software engineer at Meta’s London office, documented a grueling transition that involved 413 rejections before landing a role at Revolut. The journey highlights the structural barriers facing career switchers: the dominance of PhDs, the geographic friction of remote applications, and the harsh reality that 'big name' resumes only get you in the door, not through it.

The Infrastructure of a Career Switch

The engineer’s initial attempt to pivot while employed at Meta failed because self-study lacked the necessary pressure. Realizing the carrot wasn't working, they moved to Moscow to attend Skoltech, drawn by free tuition and proximity to family. The academic shock was immediate; the first homework assignment involved manually calculating monstrous gradients that froze the engineer for a full day. This underscores a critical infrastructure gap for SWEs: the mathematical foundations of ML are not something you can 'hack' with software engineering intuition alone.

Rejection as a Data Point

The application phase was a masterclass in resilience. After applying to roughly 100 roles in London from Moscow and receiving zero callbacks, the engineer suspected location bias was the culprit. Moving back to London improved response rates, but the interview funnel remained a bottleneck. Early failures at Isomorphic Labs, Huawei, and the UK’s AI Safety Institute were attributed to a lack of specific ML knowledge gaps. The engineer eventually landed an internship at Mistral AI, though they candidly admit the Meta brand likely secured the initial conversation, even if the actual work at Meta had zero relevance to machine learning.

The PhD Ceiling and the Revolut Breakthrough

Despite gaining experience at Mistral and publishing a paper accepted at EACL, the engineer hit a wall with frontier labs like DeepMind and Anthropic. The competition was stiff: candidates with 10 years of ML-specific experience, multiple A* conference publications, and PhDs consistently edged them out. The breakthrough finally came from Revolut, a neobank that had quietly released a foundation model for financial transactions. The engineer was hired into a small, high-momentum research team, proving that niche industrial AI labs may offer a more accessible entry point for experienced engineers than the big tech giants.

Key Takeaways

  • Brand Leverage: A 'big tech' resume (Meta) is often the only thing that gets a career switcher past HR filters, even if the prior work was unrelated to ML.
  • Geography Matters: Applications sent from Moscow were ignored; the same profile sent from London got interviews. Location bias is a real infrastructure issue in hiring.
  • The PhD Barrier: For frontier labs, 10 years of general engineering experience does not compete with 10 years of ML-specific experience and a PhD.
  • Start Small: The engineer found success at a smaller, specialized team (Revolut) rather than a massive research org, where they could take on broader scope.

The Bottom Line

The '413 No' story is a brutal reminder that AI research hiring is an elite club with high barriers to entry. For infrastructure engineers, the path isn't just about learning PyTorch; it's about navigating a hiring market that systematically undervalues generalist engineering experience in favor of specialized academic credentials.

Builder’s Advice

  • Internal Transfers are Best: If you can switch to ML infra within your current company, do it. You keep your income and avoid the 'remote applicant' penalty.
  • Network Aggressively: The Revolut job was found via a friend’s DM, not a job board. Your network is your net worth in this transition.
  • Accept the 'Idiot' Phase: The engineer describes a 'permanent feeling of being an idiot.' If you aren't feeling stupid, you aren't learning fast enough.