Talos turned a 57-page course deck into 11 study notes, 11 knowledge cards and one overview image in about 7 minutes, using a prompt that fixes granularity.
This article documents the whole process, from a single course deck in to a complete set of revision materials out: the prompt verbatim, how Talos executed it, the structure of what came back, and the steps you can copy.
Finished output and process screenshots

*01 Chat interface*

*02 File list*

*03 Knowledge cards overview*
Project background
Turning a lecturer's slides into material you can revise from directly is a fixed chore at the end of every term. For a 57-page course deck, splitting out the knowledge points, writing notes one by one, then laying them out as cards and an overview image usually takes 4-8 hours (estimate). With Talos this time, about 7 minutes produced 11 study notes, 11 knowledge cards and 1 overview image.
Challenge
Three things make the conversion from slides to revision material hard. The source file was an old .ppt binary, and its formula symbols were stored in the private encoding of the Symbol font: extracting the text directly drops every arrow, ∈ and ∀. The deck is laid out to the rhythm of the lecture, so one page may cram in several concepts while five consecutive pages cover a single idea, and the granularity of the knowledge points has to be re-split and re-merged. Notes and cards together come to more than twenty files, and their layout still has to be consistent. A general-purpose tool can give you a summary or an outline, but the output is continuous text: splitting the knowledge points and laying out the files is still manual work. The goal here was one complete set of material that can be opened and revised from straight away.
Method
Step 1: Design a structured prompt
The prompt sets out the role, the processing flow, the structure of each note, the design rules for the cards and the delivery directory in one pass. This is the full prompt used:
Role
You are a study assistant with a background in instructional design. Your task is to turn one of the lecturer's PowerPoint decks into structured notes and knowledge cards that students can revise from directly.
Input
I will provide a PPT file (PDF or PPTX format).
Processing flow (follow the steps strictly)
Step 1: Parse the deck and extract the knowledge points
Read the whole deck and identify every knowledge point in it.
Decide the granularity of each knowledge point: one knowledge point = one complete concept, method or conclusion that can be explained and revised on its own.
If one slide contains several knowledge points, split them; if several slides cover the same knowledge point, merge them.
Output a list of knowledge points in this format:
Knowledge point 1: [name]
Knowledge point 2: [name]
...
Step 2: Generate structured notes (Markdown)
Create a separate Markdown document for each knowledge point, named: 01_knowledge-point-name.md
Each Markdown file must contain the following structure:
Knowledge point name
One sentence summarising what this knowledge point is
Core content
What this knowledge point is about (in language a student can follow)
Key definitions, formulas and conclusions
If there is a derivation, keep the key steps
Why it matters
Where this knowledge point sits in the course
How it relates to the knowledge points before and after it
How it might be examined in a test or assignment
An example
1-2 concrete examples to help understanding
If the deck has worked examples, keep them and add an explanation
Common misunderstandings
Where students most often go wrong
Concepts that are easy to confuse
Self-test questions
2-3 questions you can use to test yourself (with answers)
Step 3: Generate knowledge cards (images)
Create one knowledge card image per knowledge point, named: 01_knowledge-point-name.png
Design requirements for each card:
Size: suitable for viewing on a phone (portrait, e.g. 1080x1440)
Top: the knowledge point name (large, prominent)
Middle: the core content (keywords + short sentences, no long blocks of text)
Bottom: a one-sentence summary or mnemonic
Style: clean, clear, with a distinct visual hierarchy
Colour: no more than 3 colours, avoid anything busy
Step 4: Assemble the overview image
Join all the knowledge cards in order into one overview image, named: knowledge-cards-overview.png
Layout: a grid (e.g. 3 per row)
Label each card underneath with its number and name
Add a title at the top of the overview: XX Course Knowledge Points Overview
Step 5: Organise the delivery
Arrange all files into the following structure:
study-notes/ ├── 01_knowledge-point-name.md ├── 02_knowledge-point-name.md └── ...
knowledge-cards/ ├── 01_knowledge-point-name.png ├── 02_knowledge-point-name.png └── ...
knowledge-cards-overview.png
Output requirements
All content in Chinese.
Tone: like a senior student explaining a problem to a junior, not like a textbook.
Break the knowledge points down completely; do not leave out anything in the deck.
If the deck contains images, charts or formulas, preserve them in words wherever possible.
Keep each Markdown file under 800 characters so a student can read it quickly.
Keep the text on the cards concise: no more than 100 characters per card.
Start
Please read the PPT I provide first, output the list of knowledge points, then generate the notes, the cards and the overview image in that order.
The prompt contains four key elements:
- Role and audience are defined: the output is restricted to "material students can revise from directly", with a senior-student tone.
- The granularity rule is decidable: "one knowledge point = one complete concept that can be explained and revised on its own" gives Talos a test for deciding which page to split and which pages to merge.
- File structure and naming are fixed: the naming rules for notes and cards and the delivery directory are settled up front, so the output needs no second round of tidying.
- Length and design have ceilings: notes under 800 characters, cards under 100 characters, portrait cards at 1080×1440, no more than 3 colours.
Step 2: Talos generates everything automatically
Once it receives the prompt and the file, Talos runs the following stages in order:
- Format parsing: convert the old .ppt binary into a readable presentation and PDF, 57 pages, in 11 seconds.
- Symbol restoration: map the swallowed →, ⇒, ∈, ∀ and other symbols back to normal characters using the Symbol font code table, then extract the text page by page, in 39 seconds.
- Splitting and merging knowledge points: split the 57 pages into 11 knowledge points using the granularity rule, and record the slide numbers each knowledge point maps to.
- Writing each note: write one note per knowledge point, with core content, why it matters, examples, common misunderstandings and self-test questions with answers.
- Rendering the cards: render 11 knowledge cards to the portrait 1080×1440 specification, with a consistent colour scheme.
- Assembling the overview and packaging: join the 11 cards into a 3-column grid overview, then organise the notes and cards into the directory structure and package them.
End to end, about 7 minutes.
Step 3: Output structure
The 57-page deck was split into 11 knowledge points, each with one note and one matching card.
| No. | Knowledge point | Slides |
|---|---|---|
| 01 | From natural language to grammar | 2-10 |
| 02 | The four-tuple definition of a grammar | 11, 13, 18 |
| 03 | Productions and notation conventions | 12, 14-17 |
| 04 | Derivation and reduction | 19-26 |
| 05 | Powers of derivation and the two closures | 27-29 |
| 06 | Language, sentences and sentential forms | 35-38 |
| 07 | Finding the set a grammatical category represents | 30-34 |
| 08 | Equivalent grammars | 39-41 |
| 09 | Constructing grammars by recursive definition | 42, 43, 48, 49 |
| 10 | Three construction examples: palindromes, expressions, function prototypes | 44-47, 55-57 |
| 11 | Construction difficulties: multiple characters counted together and exchange rules | 50-52, 54 |
A few details worth noting: all 11 notes stayed under 800 characters, the longest at 748. The slide numbers for each knowledge point were kept as a separate list, so you can go back to the original deck by page while revising. The only figure in the deck (a syntax tree) was re-described in words and written into the corresponding note, so nothing was lost. The worked examples from the original deck were kept as well, with explanations and answers added.
Results
| Metric | Manual | Talos | Change |
|---|---|---|---|
| Total time | 4-8 hours (estimate) | About 7 minutes | About -97% to -98% |
| Deck parsing | Page by page, copying key points by hand | 57 pages extracted automatically, format conversion in 11 seconds | — |
| Knowledge point mapping | Noted while reading, easy to miss | 57 pages split into 11 knowledge points, with a page mapping | Coverage can be checked |
| Symbols and formulas | Entered by hand | Symbol private encoding restored automatically | — |
| Output format | Laid out by hand | 11 .md files + 11 1080×1440 cards + 1 2400×4552 overview | Ready to use |
| Reusability | Starting from scratch for every course | Prompt saved and reused; just swap the deck | High |
Talos processed the 57-page deck in about 7 minutes, producing 11 study notes, 11 knowledge cards and 1 overview image, 24 files in total. The cards were rendered in portrait at 1080×1440, and the overview was laid out on a 3-column grid at 2400×4552.
Business impact
For students and training settings that have to work through a large volume of course material repeatedly, the value of this flow is that it turns a one-off tidying job into a reusable template:
- Single run: the revision material for one course drops from 4-8 hours of organising to 7 minutes, and what you save is the most pressured stretch of the term.
- Repeat runs: once the prompt is saved, a different course only needs a different deck; neither the granularity rule nor the layout specification has to be rewritten.
- Information quality: every knowledge point maps back to slide numbers, so you can locate the original slide and check it while revising, rather than mixing content up during the tidying.
- Format compatibility: the notes are Markdown and can be edited in any editor; the cards are images, ready to be sent to a phone and flipped through at any time.
Replicable takeaways
- Write the granularity rule into the prompt: giving a criterion for "what counts as one knowledge point" works far better than writing "split it a bit finer".
- Put ceilings on length and design: state that notes stay under 800 characters, cards under 100 characters and colours under 3, and the output becomes consistent.
- Fix file naming and directory structure: spell out the naming rules and the delivery directory from the start, and there is no second round of tidying once you receive the files.
- Save it as a template: keep the prompt that has been validated, replace only the deck next time, and the marginal cost is close to zero.
FAQ
Q: Can it handle old .ppt files?
A: Yes. This deck was an old .ppt binary, and Talos converted it into a readable presentation and PDF before parsing. If the formula symbols in a deck are lost, they are restored to normal characters using the font code table.
Q: Can I edit the generated notes myself?
A: Yes. The notes are output as Markdown files, so you can edit the body text and add or remove self-test questions in any editor.
Q: How are the knowledge points split, and will content be missed?
A: They are split using the granularity rule given in the prompt: one knowledge point equals one complete concept that can be explained and revised on its own. Once the split is done, each knowledge point records the slide numbers it maps to, so you can check it back page by page. This time 57 pages became 11 knowledge points, and the only figure in the deck was preserved in the notes as a written description.
Q: Can this method be used for other courses or other material?
A: Yes. The core is the combination of a granularity rule, a fixed structure and a length ceiling, which applies equally to other decks, training handouts and product documentation; you only need to adjust the criterion for a knowledge point and the card design requirements in the prompt.