Reducing the ML feed latency and video repition once user comes back

[harshita-srivastava-yral]

• Client needs to support reordering of the items. Immediately show more relevant items first (Proirity on client side)
• Usability issue - 1st 10 I get from post cache and then next set comes from personalised feed. When I come back later, I see the same feed for few scrolls.
• We store a personalised feed, first request is to personalised feed and next goes to off chain and both are generating separate feed. That gets pushed to individual canister. Real. 1 for immediate consumption and another to be store in the canister for the later use.
• We can see how can we make the fresh and relevant content.
• Benefit: We cannot refresh the canister everyday ? No, this also has a problem with relevancy for users visiting the app multiple times in a day. Also, we are refreshing for every user even when we don't require to refresh.

[Natasha-GB]

• Can be brought down to 1 or 2 secs
• Need to enable "Preview" feature for reducing further latency
• Caching will still be used to serve a relevant and fast feed

[harshita-srivastava-yral]

• Exploration on Akash require 2-3 days for testing and shortlisting the good model
• Rented a GPU node, jupiter image was smoothly running however after sometime there was no render available.
• Create issue on the akash discord or git side as this isn't the expected behaviour

[harshita-srivastava-yral]

• Theta doesn't have spectrum of GPUs like Akash.
• While Akash is fixing its thread we can keep working with Theta

[harshita-srivastava-yral]

• Work on duplication of the videos - Requires phash integrations for this and setup its pipeline