Loading page
About this tool
Build a week-by-week AI study path from prompting basics to fine-tuning, ordered by prerequisites and your hours per week.
AI Learning Roadmap Builder turns a goal, a starting point and your weekly study hours into an ordered curriculum that runs from how language models work through retrieval, evaluation and parameter-efficient fine-tuning. Modules are linked by hard prerequisites and sorted topologically, so nothing is scheduled before the thing it depends on, and each module's hour budget is divided by your available hours to place it in a specific study week. Useful for self-taught learners and for managers who need a defensible upskilling plan rather than a reading list.
Open AI Learning Roadmap Builder on AltFTool — it loads instantly in your browser.
Provide your input — an image, text, or data.
Let the tool analyze or generate the result.
Review, refine, and reuse the output wherever you need it.
A depth-first topological sort guarantees embeddings come before retrieval and transformers before LoRA.
Each module carries a study-hour estimate that assumes you build the practice project, not just watch a video.
Enter the weeks you have and the plan tells you the hours per week required to actually finish.
It depends entirely on the end point. Reaching reliable prompting takes roughly 15-25 study hours, building a retrieval application on top of model APIs adds around 60-90 hours if you already code, and fine-tuning open-weight models typically needs 150+ hours because it requires machine-learning foundations and transformer internals first.
Not for prompting, structured output or governance work, which are all covered without code. You do need Python the moment you call model APIs directly, build retrieval pipelines or fine-tune anything — the roadmap adds a 20-hour Python module automatically unless you say you already have it.
Prompting changes what you say to a fixed model, retrieval-augmented generation adds your own documents to the context at query time, and fine-tuning changes the model's weights on your data. Retrieval is the right answer for facts that change; fine-tuning is for format, tone or a task the base model handles badly even with good context.
Learn parameter-efficient tuning such as LoRA or QLoRA first. It trains a small number of adapter parameters instead of all the weights, so it runs on a single consumer GPU and gives you the whole workflow — data formatting, training run, evaluation — at a fraction of the compute cost before you decide whether full-weight training is justified.