The State of the Market
ML engineering hiring is more skills-based than pedigree-based. Unlike some engineering niches, ML teams are hungry for people who can ship — and they know that strong engineers can learn the ML-specific parts.
The gap you need to close isn't "you're not an ML person." It's "can you demonstrate you can do ML work?" Projects, open source contributions, and relevant experience matter more than ML-specific credentials.
Three Paths Into ML Engineering
Path 1: ML Platform / Infrastructure
The fastest path for most SWEs. These roles build the tools and infrastructure that ML teams use:
- Feature pipelines and feature stores
- Training infrastructure (job schedulers, distributed training)
- Model serving and inference optimization
- Experiment tracking and model registries
- Data versioning and lineage
What you need: Strong distributed systems skills + familiarity with ML workflow. No PhD required. No expectation of producing novel models.
Why it works: Companies with mature ML teams have realized that ML research productivity is bottlenecked by infrastructure. Pure SWE skills are the primary requirement.
Path 2: Applied ML / ML Engineering
Building and productionizing ML systems for specific business domains. You don't invent new algorithms — you apply existing ones well.
Examples: recommendation systems, search ranking, fraud detection, content moderation, demand forecasting.
What you need: Solid ML fundamentals, feature engineering, model deployment, A/B testing. The ability to own a model end-to-end.
The bridge: start by taking ownership of an ML component in your current role. Even instrumenting an existing model, improving its monitoring, or owning a retraining pipeline counts.
Path 3: LLM / GenAI Engineering
The newest and most accessible path for SWEs. These roles involve:
- Fine-tuning and adapting LLMs for specific tasks
- Building RAG systems and LLM applications
- Prompt engineering and evaluation
- Integrating LLMs into products
What you need: API fluency, some understanding of transformers, strong software skills for building reliable systems around non-deterministic models. This is primarily a software problem, not a research problem.
Building the Right Portfolio
The single most effective thing you can do: ship something that uses ML and is publicly visible.
What Makes a Good Portfolio Project
Not this:
- "I trained ResNet on CIFAR-10 and got 94% accuracy"
- A Jupyter notebook with exploratory analysis
- A tutorial walkthrough you followed
This:
- A deployed application using ML that solves a real problem
- An open-source contribution to an ML tool
- A published benchmark of different approaches on a real-world dataset
- A blog post that demonstrates genuine understanding (not a summary)
Project Ideas by Path
ML Infrastructure:
- Build a mini feature store with an offline/online store split
- Implement a model registry with versioning and stage promotion
- Build a training job manager that handles GPU allocation and experiment tracking
Applied ML:
- Train and deploy a model for a domain you know (finance, biology, legal, etc.)
- Build a recommendation system with proper evaluation (offline metrics + A/B test simulation)
- Reproduce a production system from a research paper (YouTube's DNN, Uber's demand forecasting)
LLM/GenAI:
- Build a RAG system over a specific document corpus and evaluate retrieval quality
- Fine-tune an LLM for a specific task and measure against prompting baselines
- Build an evaluation framework for LLM output quality
How to Position Yourself
Don't: "I'm a software engineer looking to get into ML"
Do: "I'm an engineer who builds production ML systems" + show evidence
The difference: the first is aspirational, the second is a claim that needs to be backed up. Get the evidence first, then make the claim.
The LinkedIn/Resume Reframe
Before: "Backend Engineer at Company X | Python, distributed systems, databases"
After: "ML Engineer at Company X | Built real-time feature pipeline serving 50k predictions/day | Python, MLOps, distributed systems"
The second version requires that you've actually done ML work. If you haven't yet, that's the work.
Targeting the Right Roles
Job Title Hierarchy
Entry-level ML: ML Engineer I / Junior ML Engineer
Data Scientist (at ML-heavy companies)
ML Platform Engineer
Mid-level: ML Engineer / Senior ML Engineer
Applied Scientist (Amazon)
Staff ML Engineer
Senior/specialized: Principal ML Engineer
ML Research Engineer
Research Scientist (requires publication record)
For SWE-to-ML transitions: target ML Engineer or ML Platform Engineer at mid-level, not entry-level. You have more experience than entry-level candidates — just different experience.
Company Stage Matters
Big Tech (Google, Meta, Amazon): Specialization is valued. ML Platform roles exist as distinct teams. Interview processes are structured and foregiving for strong SWEs.
Mid-size tech (Stripe, Airbnb, Lyft): Generalist ML Engineers who can do the full stack. You need both systems and ML skills. Faster to learn, more varied work.
Startups: "ML Engineer" often means "whoever does ML." Opportunity to own a lot, but less structure for learning. Works well if you're self-directed.
Consulting/agencies: Lower bar to entry, good for building diverse experience, but can be shallow.
The Preparation Stack
Month 1: Skills
- Complete one end-to-end ML project (feature engineering → training → deployment)
- Implement core algorithms from scratch (linear regression, logistic regression, simple neural net)
- Get comfortable with PyTorch basics
Month 2: Depth
- Study one domain deeply (NLP, recommendations, or fraud detection)
- Read 3-5 ML system design papers (YouTube DNN, Instagram explore, etc.)
- Contribute to an open source ML project
Month 3: Positioning + Applications
- Polish 2 portfolio projects with public repos and writeups
- Practice ML system design interviews (STAR format for behavioral, structured for design)
- Apply to 10-15 relevant roles, not 100 generic ones
Interview Preparation
Coding Rounds
Same as SWE interviews. Don't neglect this. Strong ML knowledge doesn't offset weak coding skills.
ML Fundamentals
Expect to explain:
- How gradient descent works
- Bias-variance tradeoff
- Cross-validation and why it matters
- How to handle class imbalance
- The intuition behind a handful of algorithms
You don't need to derive backpropagation from scratch. You need to understand the concepts well enough to reason about them.
ML System Design
Use the framework from our ML system design guide. Practice with: recommendation systems, fraud detection, search ranking, content moderation.
Ready to build the skills for this transition? Start with our practical ML roadmap for software engineers.