Full-stack developers have a massive advantage in AI engineering: you can already build real products, understand databases, and know how to deploy code. The gap is narrow. You need: LLM APIs + RAG + evaluation + one shipped AI feature.
4 weeks
Month 1: LLM API Fluency
- Pick an LLM provider (Anthropic or OpenAI) and commit to it for now.
- Build a basic chatbot and a summarizer — small, working, end-to-end.
- Understand tokens, context windows, streaming responses, and cost per call.
- Integrate the model into a project you already own, so you're learning on familiar ground.
8 weeks
Month 2–3: RAG End-to-End
- Add RAG to your existing project, or build a document-question-answering system from scratch.
- Embedding: turn your documents into vectors.
- Vector store: start with Qdrant or pgvector — you already know Postgres, so lean into it.
- Retrieval: top-k, metadata filtering, and handling "no good match."
- Evaluation: a small golden set to measure whether answers are actually grounded.
Lessons for this stage
- What is RAG and When Does It Beat Pure Prompting15 min
- Embeddings Explained: From Text to Vectors15 min
- Build a RAG Pipeline from Scratch1h
- Vector Databases Beyond FAISS: Metadata Filtering at Scale15 min
- Confidence Gates: When Your RAG Should Refuse to Answer45 min
- Evaluating an AI App: Why Looks Good Isn't Enough20 min
4 weeks
Month 4: Agents + FastAPI
- Build a simple agent with tool use: file search and a web call are enough to learn the loop.
- Replace your Express/Next.js API with a FastAPI Python backend for the AI layer.
- Understand streaming for LLM responses so the UX feels live, not frozen.
8 weeks
Month 5–6: Ship + Showcase
- Ship a full project combining your web skills with AI features.
- Write it up: the problem, the architecture, and the trade-offs you made.
- Update your resume with AI projects framed in outcomes, not buzzwords.
- Apply to AI engineering roles — your shipping track record is the differentiator.
Skill priorities
Learn now
- LLM API usage (Anthropic/OpenAI)
- RAG with a real document corpus
- Evaluation basics
- FastAPI for Python services
Learn soon
- Agents with tool use
- MLOps and deployment
- Observability and cost tracking
Learn later
- Fine-tuning
- Multi-agent systems
- Custom embedding models
Ignore for now
- Relearning web fundamentals you already know
- Deep ML theory (that's Type 2 engineering)
- Framework-hopping
- Rebuilding your entire stack from scratch