Carl B

@carl-b

23 Following

23 Followers

Carl B pfp

Thinking about this a bit more and considering the unifying designs mentioned by @balajis.eth the markdown comments by @vitalik.eth. Another solution could be to provide a platform where organizers can define a visual template (eg the NS presentation layout) and slides are defined in a simple markdown language. Presenters can provide a URL or upload a presentation and all the different formats are converted into markdown which makes it easier for the LLMs to parse. So basically, a theme is defined and the slides are in markdown. LLM checks for slide count, bullets per slide, titles etc. All presentations will have the same fonts and form factor. This maintains a uniform design, makes it simpler (?) to create presentation slides, and easier to validate. Thoughts?

0 reply

0 recast

0 reaction

Carl B pfp

Here's my solution for the Presentation Validator. Currenty it only handles Google Slides and works like this: - Paste a link or upload a file (uploading not implemented yet) - Check format and call the correct parser (GoogleSlidesParser) - Fetch the JSON from googleapis and feed into OpenAI to find: - Slides count - Fonts used - If it has title (not sure how to detect this from the JSON data) - Videos count - Audio count - Images count - Bullets count (doesn't count them correctly for some reason) A better solution to this might be to first convert slides into PDF and then use Puppeteer to build a html page and feed this into OpenAI. This will normalize into one format and make it easier to parse and detect titles etc. Puppeteer can also be used to verify the emberdded video has audio. - https://f159276e-4914-4e30-9ad6-801c213e611e-00-2audwnsroqpal.spock.replit.dev - https://replit.com/@CarlBarrdahl/presentation-validator?v=1

1 reply

0 recast

4 reactions

Carl B pfp

Here's my solution built in Typescript. It works like this: - Next.js App with an API calling Farcaster - Textarea to enter comma- or space-separated usernames to send to API - Lookup farcaster ID from usernames and store in Nodes array - Compare each FID for follow status (this takes a lot of time to query for many users) - For each user that follows another store in Edges array - Return Nodes and Edges to frontend and display in Network Graph - Filter graph based on Range slider input - Display Metrics for filtered graph in tables (number of edges, adjacency matrix, shortest path) Note: Some issues I wasn't able to solve in time - fetching many users takes a lot of time due to cross-referencing the follow state - after fetching from API, the slider messes up the network graph (not sure why this is happening) https://replit.com/@CarlBarrdahl/farcaster-network-graph

0 reply

0 recast

0 reaction