You don’t need a Computer Science degree to break into AI engineering. You need a portfolio that demonstrates you can ship. The bar for “shipped” is lower than people think, and the bar for “CS degree” is higher than it needs to be.
Here’s the twelve-week roadmap we recommend to applicants who don’t have a formal CS background but want to be hireable as AI engineers by the end.
Weeks 1–2: Python Fluency, Not Mastery
You need enough Python to read other people’s code, write your own functions, work with dictionaries and lists comfortably, and not be afraid of async. That’s a two-week task, not a two-year task.
Free resources:
- Codecademy’s Python 3 course (~15 hours)
- Automate The Boring Stuff With Python — chapters 1–9
- The Python tutorial on docs.python.org if you prefer reference-style learning
Goal by end of Week 2: write a small CLI tool that fetches data from an API, transforms it, and writes JSON. Ship it as a Git repo.
Weeks 3–4: The LLM API Workflow
Sign up for Anthropic’s API. Read the Tool Use docs end-to-end. Build three things:
- A wrapper that takes any function and exposes it as a tool to Claude
- A small REPL where you can chat with Claude and it can call those tools
- A “research assistant” — a CLI that takes a topic and returns a summary using a web-search tool
This is where most beginners get unstuck — not by lacking knowledge, but by hesitating to actually call the API. Spend the $20.
Weeks 5–6: RAG From Scratch
Build a Q&A system over a body of text you care about. The text matters — pick something you’ll be motivated to debug for hours. Documentation for a tool you use is a great choice.
- Chunk the text (start naive, improve later)
- Embed it with
text-embedding-3-small - Store in Chroma or LanceDB (no managed service yet)
- Build retrieve → rerank → answer pipeline
- Add RAGAS evaluation against a 20-question golden set
By the end of Week 6, you’ll have a real RAG system and a metric you can point at to say “I made it 18% better”.
Weeks 7–8: Build A Multi-Step Agent
Use LangGraph. Build something with branching logic — a customer-support triage agent, a research agent that decides when to dig deeper, an extraction agent that handles uncertain fields. The point is to show you can reason about state.
Add LangSmith or LangFuse tracing so reviewers can see your agent’s decisions. This is the single biggest signal of “this person has done it before” you can put on a portfolio.
Weeks 9–10: Ship It
Put the system behind an HTTP API (FastAPI). Containerise it with Docker. Deploy it to Modal or Replicate or a small VPS. Get a live URL.
You don’t need Kubernetes for this — but you need to prove you can take code from your laptop to something a stranger can hit. The deployment step is what 80% of bootcamp grads still cannot do.
Weeks 11–12: Open Source + Public Build
Two things, in parallel:
- Contribute to one open-source AI project. LangChain, LangGraph, LlamaIndex, Chroma — all have “good first issue” tags. Read the contributing guide. Fix something small. Get the PR merged. This single PR will get you in front of more hiring managers than your CV will.
- Write three blog posts explaining what you built and what you learned. Substack, dev.to, Medium — anywhere with a feed. Hiring managers Google your name.
What You’ll Have At Week 12
A GitHub repo with a deployed RAG-based agent, an OSS PR in your name, three blog posts, and the vocabulary to talk about it. That’s enough to walk into any AI engineering interview in Colombo and not be the least-prepared person in the room.
The Agentic AI Bootcamp does this in 16 weeks with practitioner instructors, real client projects, and Demo Day in front of hiring partners. But if you’d rather self-direct, the playbook above will get you most of the way.