All Roadmaps

Generative AI Engineering

Building products on top of language models: retrieval, evaluation, cost, safety, and everything that decides whether the demo survives real users.

32h 30m total·6 sections·24 topics
Filter:
Sort:

Make this roadmap your plan

Sign in to save your routine, streaks, notes, and progress across devices.

0 of 24 topics completed(est. 32h 30m remaining)
0%
1

Foundations

The minimum you need to reason about a model rather than prompt it superstitiously.

5h 15m0/4
0%

Python for AI work

beginner1h 30m

Typed, testable Python with async I/O and virtual environments. Most AI engineering is ordinary engineering around a network call.

Subtopics (0/4)
Type hints
Async I/O
Dependency management
Testing I/O-bound code

How a language model produces text

beginner1h 30m

Embeddings and vector similarity

beginner1h 15m

Choosing a model

beginner1h

2

Prompting and structured output

Getting reliable, parseable behaviour out of a probabilistic system.

5h 15m0/4
0%

Prompt and context engineering

beginner1h 30m

Structured output

intermediate1h 15m

Tool use and function calling

intermediate1h 30m

Treating prompts as code

intermediate1h

3

Retrieval-augmented generation

The default architecture for grounding a model in your data — and the one most often built badly.

7h0/5
0%

Chunking and ingestion

intermediate1h 15m

Vector stores and indexes

intermediate1h 30m

Grounding and citations

advanced1h 15m

Evaluating a RAG system

advanced1h 30m

4

Evaluation and observability

The discipline that separates a product from a demo.

5h 15m0/4
0%

Building an evaluation set

intermediate1h 30m

LLM-as-judge, and its limits

advanced1h 15m

Regression testing prompts and models

advanced1h 15m

Observability for LLM features

advanced1h 15m

5

Production concerns

Cost, latency, safety and the failure modes that only appear with real users.

5h 15m0/4
0%

Cost and latency engineering

advanced1h 30m

Reliability around a probabilistic dependency

advanced1h 15m

Safety and abuse resistance

advanced1h 30m

Privacy and data handling

advanced1h

6

Going further

The techniques worth reaching for once retrieval and evaluation are solid.

4h 30m0/3
0%

Fine-tuning and when not to

advanced1h 30m

Agentic systems

advanced1h 45m

Multimodal inputs

advanced1h 15m

Generative AI Engineering | AlgoDrill