Two Reasons:
- To test a concept where I create content that could be shared with wider teams for education purposes - I wanted to build a variety of content to keep it interesting (and challenging)
- Continue to learn and develop my skills across AI/LLM tooling. I've been using Anthropic's Claude to build Web Infra and demo Applications on AWS/Microsoft for a while now; this was a different challenge
- There are definitely some minor errors in the content but I was happy to live with those and explain onwards. With more time I would have continued to refine the initial Master Research document to ensure it was 100% accurate (or as close as possible)
- It still took a decent chunk of time to create all this content even with the AI tooling - granted, a LOT less time than if I had done all this manually - I'd guess by at least a factor of 10!
- We're still nowhere near a stage where things work in one-hit; my Claude Project has 11 separate chats for Research and Quiz cration
Research:
- A combination of Anthropic's Claude Opus 4.0 and Google Gemini 2.5 Pro
- Both were set to run with Deep Reseach enabled to create initial drafts from the same prompt
- I got Claude to review Gemini's and incorporate content
- I then got Gemini to review Claude's initial combined draft and to make suggestions / example updates
- Claude then created a final version based on this feedback in tandem with mine
Quiz:
- The quiz was predmoninantly created with Anthropic's Claude
- Opus 4.0 was used to brainstorm the approach, help to create content and the initial draft website
- Sonnet 4.0 was used to help fix errors and refine to final version
Podcast:
- I used Google's NotebookLM for this
- Approach was simply to take the final Research document and give it to NotebookLM. No additional direction beyond short / medium lenght output
Slides:
- I tested both Gemini and Claude to build the outline of the slide decks however neither could generate the actual content and I didn't want to have to make it.
- Opted for Gamma.app which did a great job first time. I also got it to test the versions that Gemini and Claude suggested but preferred the native approach. Interestingly Gamma is a Claude cutomer.
Video:
- I generated the talk track for each slide using Claude Opus 4.0. At first I was going for 15 minutes but this needed to be reduced to under 10 minutes due to cost. The initial draft was good, and only needed minor tweaks
- I used Synthesia to create the actual video. I thougt I could change the Avatar after doing this test run but I'd used up my minutes and it's an expensive tool! Generating vitual me will have to wait to another time
Final Page:
- I once again used Claude Sonnet 4.0 to design and create the final page that encompasses all elements; there were some bugs it struggled with so Opus 4.0 was used to complete it. Essentially this element was just expanding the quiz page but I kept them separate in the repo