Rohan Adwankar

im learning about datacenters @nvidia and recreationally programming devtools

[Rohan Adwankar @rohanadwankar]

Currently at #2 on YC's @hackernews :O

[Rohan Adwankar @rohanadwankar]

To see how we continue optimizing the performance of how agents running on AI Factories optimize the performance of AI Factories, follow along on Github: https://t.co/J1V11xAKUa

[Rohan Adwankar @rohanadwankar]

The reason I chose this example to demo is that as we grow the performance ceilings we unlock new configuration options / optimizations. In the past this may have been seen as complexity but with agents, the datacenter can recursively self improve to customize to your needs.

[NVIDIA AI Infrastructure @NVIDIAAIInfra]

We developed an agent to help streamline AI factory operations management.

The AI Factory Operations Agent uses specialized agents to detect failures and correlate root causes across hardware, telemetry, Kubernetes, and more, all from one chat interface.

[Rohan Adwankar @rohanadwankar]

@NVIDIAAIInfra Check it out on Github: https://t.co/J1V11xAKUa

[Rohan Adwankar @rohanadwankar]

@NVIDIAAIInfra 🚀🚀🚀

[Rohan Adwankar @rohanadwankar]

woah thanks for the shout out! Muse and Jina are both fantastic!!

[Elastic @elastic]

Meta's Muse runs a 37.8M-param reranker on 2 CPU cores.

@rohanadwankar tore down the Muse agent VM. In its memory stack: jina-reranker-v1-turbo-en, 153 MB on disk.

Muse searches its users' memories in Postgres with vectors plus keywords. The Jina reranker orders those hits before the agent reads them.

Per the teardown, retrieval runs inside the VM, so memory chunks stay on the encrypted volume.

It's our English-only reranker from 2024. It still earns its place on a box with no GPU.

You can also run our latest @JinaAI_ models with Jina On-Prem. For self-contained and air-gapped deploys, with no outbound calls. .

Rohan's full teardown: https://t.co/W18dgq9Nt8

[Rohan Adwankar @rohanadwankar]

@jaredsisk thanks!

[Rohan Adwankar @rohanadwankar]

tried out instinct it was the best ive seen! fun fact with ws-term you can pull your agent’s s3 bucket and read its git repo to see what it really thinks about you 😉

[Rohan Adwankar @rohanadwankar]

@AniC_dev @upstash the world if @SemiAnalysis_ did InferenceX but for sandboxes so we could finally get standardized comparisons proven in CI

[Rohan Adwankar @rohanadwankar]

cant wait for when it can speak english through adhoc w.writeframes(_) instead of TTS lol

[Rohan Adwankar @rohanadwankar]

asked claude to make a teaser for my repo and it made the audio effects using just 100 lines of numpy and wave

[Rohan Adwankar @rohanadwankar]

@koao90 i gotchu https://t.co/I9SDXOiIXU

[Rohan Adwankar @rohanadwankar]

might be the funniest comment someone has ever said about something i made. i love hacker news

[Rohan Adwankar @rohanadwankar]

@AI_Pragmata yup i wrote a followup on muse as well, its up on my blog if you’re curious!

[Rohan Adwankar @rohanadwankar]

@inventcures @noahrshinn coolest thing was probably one time i asked for something and it didnt get it the first time, then 24 hours later unprompted it comes back with a demo video. that was very impressive feels like were moving beyond chatbots to agents with great memory systems

[Rohan Adwankar @rohanadwankar]

if ur curious i put up a write up on hacker news https://t.co/CdaFeVxY5m

[Rohan Adwankar @rohanadwankar]

fyi claude gives free vms for their mobile app. and you can ssh into it and run deepseek nano on @opencode. probably the easiest mobile multiagent setup

[Rohan Adwankar @rohanadwankar]

@kennethlau12393 congrats on the launch! pre-AI triage does the mutation algorithm itself have a gen-ai component?

[Rohan Adwankar @rohanadwankar]

@growing_daniel if ur complaining about oss and not putting up a patch then u dont really care about what you're complaining about

[Rohan Adwankar @rohanadwankar]

https://t.co/q1gi69T33E

[Rohan Adwankar @rohanadwankar]

made snake, except every square has a number and negative numbers shrink you. this makes solving it harder because the normal hamiltonian path solution gets more complex as the tail shrinks and you need enough room to make up for it