The 9-Month Roadmap From Data Scientist to AI Engineer
(Phase by Phase, Resource by Resource, Skip List Included)
This is Part 5 of a 5-part series adapted from How to Break Into AI Engineering in 2026. Part 1 covered the career case. Part 2 covered the 5-track curriculum. Parts 3 and 4 covered the Software Engineer and ML Engineer paths.
The complete guide (all tracks, skip lists, roadmaps, and 80+ linked resources) is available as a free PDF: How to Break Into AI Engineering in 2026
You understand models better than most AI engineers ever will. You think in distributions, significance, and experimental design. A genuine advantage. But your gap is real: you’ve been building in notebooks. AI engineering happens in production. The distance between a working notebook and a working service is the distance this path covers.
You already have: models, statistics, experimentation.
Your gap: production engineering first (APIs, containers, CI/CD), then the AI stack on top. The mental model shift matters most: from “model quality” to “system quality, including the model.” The longest path because the production gap is wider than most data scientists expect.
Three mistakes data scientists make before they even start
You stay in the notebook too long. This is the single most common failure mode for data scientists transitioning to AI engineering. The notebook is comfortable. It’s where you think. But it’s not production. Every week you spend iterating in a notebook instead of learning to deploy is a week you’re building skills that don’t transfer to the role you want. Force yourself out of Jupyter by month two. The discomfort is the point.
You optimize model quality at the expense of system quality. A model that’s 95% accurate in your notebook and crashes under concurrent requests is worse than one that’s 90% accurate and stays up. Production AI engineering is about the system, not just the model inside it. Latency, error handling, monitoring, cost management. Your transition starts the moment you accept that model accuracy is one metric among many, not the only one that matters.
You skip the production fundamentals because they feel beneath you. Docker, FastAPI, CI/CD, Git branching. You already have a PhD or years of modeling experience. Learning containerization feels like a step backward. It’s not. It’s the step that makes everything else possible. The data scientists who skip months 1–2 of the roadmap hit a wall at month 4: they can build an agent that works on their laptop but can’t deploy it, can’t monitor it, and can’t hand it to another engineer to maintain. Every shortcut in months 1–2 becomes a blocker in months 5–6.
Months 1-2: Learn to ship.
Before you touch anything AI-specific, close the production gap. This is the phase most data scientists want to skip. Don't. Everything after this phase depends on it.
Week 1: FastAPI + Docker. Week 2: Git + first deploy. Weeks 3-4: MLOps Zoomcamp 1-4.
If you already deploy models via APIs and containers, skim this phase and use the checkpoint to confirm you can move on. If "deploy" means "share a notebook link," this phase is non-negotiable.
- FastAPI: Build one endpoint that takes a JSON payload and returns a prediction. Don't learn the whole framework. You need POST routes, Pydantic models, and async. That's it.
- Docker: One Dockerfile that runs your FastAPI app. Learn docker build, docker run, docker compose up. Skip Kubernetes entirely at this stage.
- Git: Branch, PR, merge. If you're still committing to main, fix that before anything else.
- DataTalksClub's MLOps Zoomcamp (16K stars): Weeks 1 through 4 only. Experiment tracking, pipelines, deployment. Skip the Kubernetes and advanced monitoring sections until you've shipped something real.
By end of month 2, deploy a sentiment analysis API that takes a text payload via FastAPI, runs inference, returns structured JSON, and runs inside a Docker container on a cloud VM. That's your proof you can ship.
This phase will feel slow. You already know Python. You already understand ML. Now you're learning Docker. But this is the phase that separates data scientists who build AI products from data scientists who demo them.
The bigger shift isn't the tools, it's the mindset. In data science, you run an experiment, analyze results, and present findings. In production AI engineering, your code responds to events in real time, handles concurrent users, and fails gracefully when inputs are unexpected. Start thinking in requests and error codes, not notebooks and cell outputs.
The checkpoint: Deploy a simple API endpoint inside a Docker container. If you can hit it from curl and get a response, move on. If "deploy" still means "share a notebook link," you're not done.
Months 3-4: Add the AI layer.
Now you're ready. Start with Anthropic's API quickstart or OpenAI's quickstart. You already understand embeddings mathematically, which puts you ahead of every software engineer in this phase.
- DAIR.AI's Prompt Engineering Guide (80K stars) for prompt engineering.
- NirDiamant/RAG_Techniques (30K stars) for patterns.
- openai/openai-cookbook for production-tested recipes.
- Pinecone or Qdrant for the vector database.
The mental model shift happens here. Your RAG pipeline will return perfect answers on your test set and hallucinate on the first real user query. The instinct is to improve the model. The fix is usually in the chunking, the retrieval, or the prompt. Debug the system, not just the model.
Get Paolo Perrone’s stories in your inbox
Join Medium for free to get updates from this writer.
The checkpoint: A RAG system deployed as an API, containerized, with logging and basic monitoring. If someone on your team can hit the endpoint and get a useful answer without you being in the room, you're ready.
- pguso/rag-from-scratch builds RAG from scratch with no frameworks. Especially useful for data scientists because it forces you to build the system, not just the model.
- patchy631/ai-engineering-hub for step-by-step RAG project tutorials.
Like what you're reading? The complete guide includes the full 5-track curriculum with skip lists, checkpoints, and 80+ ranked resources. Get the full PDF.
Months 5-7: Build the system.
Agents:
- Anthropic's tool use docs or OpenAI's function calling guide.
- huggingface/agents-course or panaversity/learn-agentic-ai (4.4K stars).
- LangChain Academy for LangGraph.
Data scientist milestone: Build a RAG pipeline with an A/B evaluation framework: two chunking strategies, measured with RAGAS faithfulness scores, statistical significance tested. This is where your experimentation background becomes your edge.
Now deploy your advantage. This is where your statistical background stops being irrelevant and starts being rare.
Evaluation:
You already think in metrics, distributions, and significance testing. Most AI engineers don't. Apply that thinking to LLM outputs: faithfulness scores, answer relevancy distributions, regression testing across prompt versions. This is your differentiator. Lead with it in every interview.
The checkpoint: An agent that chains multiple tools, with a full eval suite that tracks performance across prompt versions. If you can show the eval dashboard and explain what the metrics mean, you have something most AI engineering candidates can't match.
- jamwithai/production-agentic-rag-course for a complete agentic RAG project with monitoring and deployment.
- NirDiamant/GenAI_Agents (25K stars) for 64 agent notebook implementations as supplementary project material.
Months 8-9: Ship and break.
Deploy an AI feature that handles real traffic.
- anthropics/claude-quickstarts for deployable reference projects.
- Langfuse or Helicone for monitoring.
- microsoft/PyRIT (4.8K stars) for red-teaming before you ship.
- confident-ai/deepteam (3.1K stars) for the offensive playbook: 50+ vulnerabilities mapped to OWASP and NIST.
Ship it. Monitor it. Break it. Fix it. That cycle is what separates an AI engineer from a data scientist who knows some LLM APIs.
If your deployment involves self-hosted models or your team is hitting cost or latency ceilings with hosted APIs, circle back to Track 4 (Hardware & Inference) in the full guide. The serving and optimization skills there become directly relevant once you're operating a real system under real constraints.
The checkpoint: A deployed system with an eval dashboard that proves it works. That's your portfolio piece. Nothing else comes close in an interview.
- alexeygrigorev/ai-engineering-field-guide for real interview take-home assignments. Build one as your final portfolio piece.
Start interviewing at month five or six for eval-focused roles. Target full-stack AI engineer positions at month seven or eight when your portfolio is stronger. Lead with the eval pipeline. Most candidates talk about what they built. You can show how yours works.
Start today: install Docker. Put a FastAPI endpoint in a container. Hit it from curl.
This article is adapted from How to Break Into AI Engineering in 2026. The complete version includes all three transition roadmaps (SWE, ML, DS), 5 learning tracks with skip lists and checkpoints, and 80+ linked resources.
Coming from software engineering or ML engineering? The SWE (6 months) and ML Engineer (8 weeks) roadmaps are in Parts 3 and 4.
