Building a research prompt with the Syllabus Reverse-Engineer, running it in the AI assistant you already use, and checking what comes back before you commit months to it. Screenshots sit behind the same "More detailed info" toggles the other field guides use.
Short version: your own assistant does the research, so the tool stays free, private, and personal to you. The trade-off is that you check the links yourself.
Mapping a university course to free resources takes real research: finding the current syllabus, then hunting down lectures, notes, and problem sets for every week. An AI assistant with web search can do that in a few minutes, but only if it is asked the right way. That is the part this tool handles. It turns a course name and a few facts about you into a detailed prompt with strict rules about sources, honest gaps, and output format.
Because the research runs in your assistant, nothing you type is sent to Deliberate Learners, there is no account, and the plan is fitted to your background and your weekly hours rather than a generic version of the course. The cost is that quality depends on your assistant and on web search being switched on, and nothing here can check links automatically. Part 4 covers how to do that yourself.
Build your prompt. Everything below happens inside syllabus-reverse-engineer.html, entirely in your browser.
One form, top to bottom: what you want to learn, who you are, and how you like to study.
Only the course (or goal) field is required. Everything else has a sensible default. The tool remembers what you typed in this browser, so you can close the tab and come back later. The ↺ button at the top right clears the form, and ๐ switches between light and dark mode.
Name a course you already know about, or describe where you want to end up and let the assistant pick the courses.
A specific course produces one detailed plan for that course.
A learning goal first asks the assistant for a sequence of three to
six courses from top universities that reach the goal, in prerequisite order. It then
builds the full plan for the first course only, so the answer doesn't get cut off.
Reply next course in the same chat to get the next one.
The chips under the field are examples. Click one to fill the field.
University, course code, and title beat a topic. MIT 6.006 Introduction to Algorithms, not algorithms.
The first thing the assistant does is search for the course's real syllabus. A course code gives it something unambiguous to search for. "Machine learning" matches hundreds of courses, while "Stanford CS229 Machine Learning" matches one. If you don't know the code, the university and the full course title usually work too.
For goals, describe the outcome rather than a topic list: "understand how operating systems work" or "get to graduate-level linear algebra". The assistant then works backwards to the courses.
Optional, but it is what makes the plan yours: the assistant uses it to flag what you can skim and what to review first.
A sentence or two is enough. Be specific about the maths and programming you already have, and about what's rusty. That tells the assistant which prerequisite refreshers to add and which early weeks you can move through quickly. If you leave it blank, the prompt asks the assistant to assume the course's official prerequisites and to tell you what they are.
Hours set the pace; the assistant choice changes the instructions and the open button; the checkboxes shape which resources get priority.
Hours per week (1 to 60) is what the assistant uses to estimate how many calendar weeks each course week will take you. A university week assumes a full-time student, so at six hours a week one course week often stretches over two.
Your AI assistant doesn't change the prompt itself. It changes the "Before you paste" instructions (each assistant turns on web search differently) and which site the Copy & open button opens.
I learn best with tells the assistant which kinds of material to favor. Keep Problem sets ticked if you can: working the problems is what separates taking a course from watching one. Include free-audit courses allows Coursera and edX courses you can audit for free (labeled "free audit" in the answer). Untick it if you only want fully open material.
Build my prompt shows the full text. Copy it, or use Copy & open to copy and jump straight to your assistant.
The prompt appears in a read-only box below the form. Read it once. It's plain English, and seeing what the assistant is asked to do makes its answer easier to judge. Copy prompt puts it on your clipboard. Copy & open copies it and opens your assistant in a new tab. For Claude, ChatGPT, and Perplexity it also tries to pre-fill the chat box. If the box comes up empty, the prompt is already on your clipboard, so just paste it. Gemini opens without pre-filling. With "Other", the button simply copies again.
Under the buttons, Before you paste lists what to do next, including how to turn on web search in the assistant you picked:
Run it in your assistant. The only hard requirement is that the assistant can search the web.
Without search, an assistant can only guess at links from memory, and guessed links are how study plans end up full of dead pages.
| Assistant | What to do |
|---|---|
| Claude | Make sure web search is turned on before you send the prompt. |
| ChatGPT | Turn on Search, or use a model that can browse. |
| Gemini | Searches Google on its own. Deep Research mode gives a more thorough result. |
| Perplexity | Searches by default. Research mode is the most thorough. |
Assistants rename and move these settings often, so if you can't find the switch, ask the assistant itself how to turn on web search. The prompt also tells it to say so at the very top if it can't search. If you see that warning, stop and fix search before using the plan.
Don't trim it. The link rules and the output format near the end are what keep the answer honest and usable.
A careful answer runs many searches. A slow answer usually means real research, not a stall.
In goal mode, reply next course for each course in the sequence. In either mode, follow-ups refine the plan in place.
Follow-ups that tend to work well in the same chat:
Week 5 is marked partial. Search again for a free resource that covers what's missing.Replace the free-audit courses with fully open material.I only have 3 hours a week now. Re-pace the plan.Give me the week-by-week checklist again as a single Markdown list I can paste into my notes.Read what comes back. The prompt asks for the same seven sections every time, in this order.
| Section | What to look for |
|---|---|
| 1. Summary | The syllabus term it found (e.g. "Autumn 2018") and a link to it. An older term is normal, since many courses publish recordings from one year. |
| 2. Prerequisite check | What the course assumes, with a free refresher for anything your background suggests is missing. |
| 3. Coverage map | One row per week with its free resources and a coverage rating. Scan this first. |
| 4. Week-by-week checklist | - [ ] items for every lecture, reading, and problem set, each with a link and a time estimate. |
| 5. Projects | Free equivalents of the course projects, and a place to publish your work as proof you did it. |
| 6. Gaps and caveats | What free material can't replace, such as graded proofs or instructor feedback. |
| 7. Links to verify | Every URL once, as a list you can click through in Part 4. |
The ratings are the assistant's judgment of rigor, not just topic overlap.
A plan with a few partials is normal and honest. A plan where every week is "full" deserves a closer look at the links.
Check the plan before you commit to it. A few minutes here saves a month of following a broken plan.
The prompt tells the assistant to cite only pages it found through search, which cuts down made-up links but doesn't eliminate them. Assistants can still mislabel a real page or point to an old version.
Match the symptom to the likely cause before starting over.
| Symptom | Likely cause and fix |
|---|---|
| Answer opens by saying it can't browse | Web search is off. Turn it on (Part 2, step 01) and send the prompt again in a new chat. |
| Several links are dead | Courses move between terms. Ask it to search again for those weeks, or look for the same course on MIT OpenCourseWare or an archived offering. |
| The answer stops partway through the weeks | It hit its length limit. Reply continue from week N. |
| The plan feels generic, not fitted to you | The background field was empty or vague. Add specifics and rebuild the prompt. |
| Copy & open opened an empty chat | That site didn't accept the pre-fill. The prompt is already on your clipboard, so paste it. |
| It picked the wrong course | The name was ambiguous. Add the university and course code, then rebuild. |