Vineeth Kalluru

teaching machines to think, trying to think like one myself⚡

Applied AI @nvidia · ex google TPU, sambanova. obsessively building things nobody asked for 🤷🏻‍♂️

[Vineeth Kalluru @vineethkalluru]

I finally pursued my long time side project idea: turned my phone into a scratchpad for Codex! ✍️ → 🤖

Meet dev.board:
An open-source Android app that lets you sketch an idea, architecture diagram, UI flow, or handwritten note and send it directly to Codex/Claude code over your local Wi-Fi.

No cloud upload. No manual file transfer.

Just:
Draw with a stylus or your finger
Tap Send
Use the codex skill to read your latest scratchpad. Simple.

I built this because keyboard and voice aren’t always the best ways to communicate an idea. Sometimes the fastest interface is still a blank canvas.

The project includes:
- An Android scratchpad with a dotted, expandable canvas
- Local Wi-Fi discovery and pairing
- Direct image transfer to your Mac
- A Codex skill for retrieving the latest sketch

It started as an experiment on how to use my Samsung Ultra S-Pen effectively and evolved into a more general idea: your phone/ipad can be a visual input device for an AI coding agent.

I know we have apps like tl;draw, excalidraw etc. but they felt too bloated for my need.

The repo is open source, feedback and contributions are welcome:

Looking forward to hearing your feedback!

[Vineeth Kalluru @vineethkalluru]

NVIDIA will keep it compute agnostic. PERIOD.
Jensen doesn't believe in playing 'below-the-belt' games by locking other vendors out of the platform. If anything NVIDIA's infra and capital will infuse more life into the platform. If not NVIDIA then who? Obviously no one from the software world (Google, Amazon, Meta etc.) will have the incentive to keep it open. No one from the hardware players have the money to buy HF. So NVIDIA remains the only option. NVIDIA capital was necessary to keep the world's best open source platform - HF - alive

[clem 🤗 @ClementDelangue]

Super happy to share our intention to join forces with NVIDIA in a $12,930,300,000 acquisition 💛💚

10 years after starting Hugging Face, open-source AI is at an inflection point. Thanks to the community, we’ve shown that it can be a complement, and even an alternative, to closed-source APIs. But for it to happen at larger scale, it needs more compute, more support, more collaboration and more visibility. That’s why we went to talk to Jensen, who offered to do exactly that with us.

In addition to doubling down on NVIDIA’s massive contributions to open-source AI (I called them the “King of American open-source AI” earlier this year), they’ve committed to strongly supporting Hugging Face and our mission while keeping the platform open, independent and compute agnostic. The founders and the team are all staying to keep pushing this mission forward.

Together, we think we can make open source the default way to build AI, with the goal of empowering 100 million AI builders to own their intelligence rather than rent it.

Excited about the next 10 years! 🤗🤗🤗

[clem 🤗 @ClementDelangue]

Super happy to share our intention to join forces with NVIDIA in a $12,930,300,000 acquisition 💛💚

10 years after starting Hugging Face, open-source AI is at an inflection point. Thanks to the community, we’ve shown that it can be a complement, and even an alternative, to closed-source APIs. But for it to happen at larger scale, it needs more compute, more support, more collaboration and more visibility. That’s why we went to talk to Jensen, who offered to do exactly that with us.

In addition to doubling down on NVIDIA’s massive contributions to open-source AI (I called them the “King of American open-source AI” earlier this year), they’ve committed to strongly supporting Hugging Face and our mission while keeping the platform open, independent and compute agnostic. The founders and the team are all staying to keep pushing this mission forward.

Together, we think we can make open source the default way to build AI, with the goal of empowering 100 million AI builders to own their intelligence rather than rent it.

Excited about the next 10 years! 🤗🤗🤗

[Vineeth Kalluru @vineethkalluru]

@BarathAnandan7 @finkd @opencode This is huge!

[FluidVoice @FluidVoiceApp]

Google is coming for us 😭

Google just dropped a tiny AI dictation model.

It runs entirely on your device.
Polishes your speech like the cloud models.
And costs $0.

Wait.

THAT’S US.😭😭

Shipping soon. Sneak peek below :)

p.s. love you @googlegemma for powering our fluid models

[Vineeth Kalluru @vineethkalluru]

@gowthami_s at that rate..fine tuning is done by support engineers and Applied AI is done by the agent itself. Which is kinda true now that I think about it 😆

[Vineeth Kalluru @vineethkalluru]

@deviparikh Connecting this to a local voice AI agent can be game changing. Almost like Jarvis.

[alphaXiv @askalphaxiv]

arXiv -> quickarXiv

Papers are often written in convoluted language that is hard to understand

Swap arXiv -> quickarXiv on any paper URL to get an instant blog with figures, insights, and explanations. Now extracted with @Zai_org's GLM OCR 🚀

Also includes code + author tweet

[FluidVoice @FluidVoiceApp]

Cloud AI dictation subscriptions are starting to look like a scam.

We just got accurate, formatted AI dictation running fully ON YOUR iPhone.

If we can do this for free, explain to me why you’re paying $10–20 every month.

[Vineeth Kalluru @vineethkalluru]

Wait.. this is interesting. Why didn't I not think of this!!!

[alphaXiv @askalphaxiv]

If doomscrolling X is part of your research workflow, we built something for you.

Introducing Paperscrolling 🚀

Get the most trending research with key ideas, figures, and audio explanations from alphaXiv Briefs

[Yaowei Zheng @code_hiyouga]

If I had to recommend a single course for getting up to speed on the current state of AI agents, it would be UC Berkeley RDI's Agentic AI MOOC. It's the third installment in Berkeley's LLM Agents course series (following LLM Agents in Fall 2024 and Advanced LLM Agents in Spring 2025), taught by Dawn Song and Xinyun Chen, with all lecture recordings and slides publicly available.

The speaker lineup spans every layer of the agent ecosystem. Twelve lectures, organized into four phases:

System design: Yann Dubois (OpenAI) on LLM agents overview; Yangqing Jia (VP at NVIDIA) on the evolution of system design from an AI engineer's perspective. Training & evaluation: Jiantao Jiao (NVIDIA) on post-training verifiable agents; Weizhu Chen (Microsoft) on lessons from training agentic models; Noam Brown (OpenAI) on multi-agent AI; Sida Wang (Meta) on statistical noise in LLM evaluations.
Real-world applications: James Zou (Stanford) on AI agents automating scientific discovery; Clay Bavor (co-founder of Sierra) on deploying agents in production; Oriol Vinyals (Google DeepMind) on multi-agent systems.
Embodiment & safety: Peter Stone (UT Austin / Sony AI) on embodied agents; Dawn Song closing with agentic AI safety and security.

A complete learning loop: a 32,000+ member global community with an active Discord, quizzes for every lecture, and a four-tier completion certificate — the higher tiers require submitting a project to the AgentX-AgentBeats competition.

Short on time? Prioritize four lectures: Clay Bavor (deployment) → Jiantao Jiao (post-training) → Yangqing Jia (system design) → Agent Evaluation. Together, they cover the full pipeline: design, train, evaluate, ship.
Course page: https://t.co/so1ydDAhCb

[Vineeth Kalluru @vineethkalluru]

@themathgeek13 Haha.. good one. I hope meownomics is working out for you

[Vineeth Kalluru @vineethkalluru]

@BarathAnandan7 @thsottiaux I don't know if you have used their voice model in Claude app. That one needs a serious upgrade!

[Vineeth Kalluru @vineethkalluru]

@BarathAnandan7 All those GPUs they bought in the last two years is paying off now. With Cursor also on their side they can be serious competitor in the coding agent space.

It's amazing to see GPUs give them serious scale to the level they can undercut competition by 2-3x

[Vineeth Kalluru @vineethkalluru]

@FluidVoiceApp @WisprFlow #myvoicemyright

[FluidVoice @FluidVoiceApp]

This swift app has 9500 ⭐️ on github

For their 10,000 star milestone,

They are planning to launch a new local dictation clean up model:

FLUID-1 MINI that can clean and format text as lists, emails , understand your intention and more!

- Takes only ~1GB of RAM
- Cloud level dictation clean up running locally
- Runs at up to 400 Tok/s
- supports almost any macbook in existence

PS: It’s us - @FluidVoiceApp

We’re here to raise the bar of what local hardware can do and we’re not stopping.

If you’re reading this and did not star it yet, what are you even doing????

https://t.co/YOc8EeD8h0

[FluidVoice @FluidVoiceApp]

The fastest free and local dictation app is on Windows now

As part of this early release,

FluidVoice windows will have an exclusive access to a research preview of Fluid-1 mini!

- 1.5 GB
- optimized to run near instant speed
- formats lists, emails, understands intent and more, fixes corrections and a lot more!
- $0

All for free on your Windows device - no compromises

Drop a comment , and we'll send you a DM to try it out!!

[Vineeth Kalluru @vineethkalluru]

@IsaacKing314 and added in Claude.md about adding this feature, removing it and detailed all the reasons why we did so. 🤷🏻‍♂️

[Vineeth Kalluru @vineethkalluru]

Maybe we can do this with - https://t.co/l2Pl0hlj1E

The idea sounds promising. But having developed compilers in the past for LLM inferencing, I have come to know that users prefer quick out of box guaranteed execution even if it's painful CLI arguments than having to figure out the implementation themselves.

I think that the audience for this are quite small. However good examples/tutorials might help.

[FluidVoice @FluidVoiceApp]

We’re officially @FluidVoiceApp

A lot of them complained about not being able to find us on X and we decided to fix it 😉

Thanks for all the love and support every single day. FluidVoice will not exist if not for the community.

Working hard every day to deliver every single thing that we promised and more!

For people who are seeing us for the first time:

We’re the fan Favorite and Fastest AI Dictation app that does local Polish on-device!

Download and waitlist links below!

[Vineeth Kalluru @vineethkalluru]

If you have a local AI model for Speech to text why not use it to record meeting notes? It makes sense if you think about it

I have been saying for sometime now that Wisprflow and Granola can be a single product and this open source repo finally did it <3

[FluidVoice @FluidVoiceApp]

I heard you all like meeting notes..

Coming soon to FluidVoice 😉

Local. Free. Private