and we're done with phase 2 already.
far smoother than phase 1, hence it was so quick
before this, you could upload a video, and that was it. the system did some stuff to it, but it really did not know what to do next
now it does:
> it reads each video and figures out what its about
> it remembers that so it can match videos to each other later
> when the model learns smth new, it updates its understanding of every video automatically
in simple terms, the system can now tell similar videos apart from different ones.
this is your "you might also like" thing.
only up from here :)
and we're moving already. phase 0 donzo:
> docker stuff is up (kafka kraft, qdrant, redis, and more)
> protobuf compiled to python + go
> seed script populates everything: videos, users, and all the stores
> some more stuff for future me
more to come :)
absolute banger read
finally understand prefill, decode, kv cache, batching, cache affinity and all the other llm fancy words
would've bought you a coffee but i'm broke
feel free to fund my nicotine addiction instead
nah, but genuinely 10/10 read