AI Architect & Builder - Inference, LLMs, SLMs GSoC Org Admin - Jenkins | Docker Captain | CNCF Ambassador
New Delhi
Joined December 2018
- Tweets7.2K
- Following1.3K
- Followers8.4K
- Likes37.7K
What an amazing day! Presented a poster at @PyTorch Conference: how we optimised the RLM paper with pre-fix caching & batched Sub calls with @vllm_project
And a full house for our Pytorch Conference talk on running LLMs using Executorch on Android!
Shivay Lamba retweeted
New research: We developed JOLT, a post-training method where one model acts as both teacher and student, with the teacher trained to give feedback the student can use.
On MATH, it matched GRPO's accuracy with 6.5x fewer training tokens.
perplexity.ai/hub/blog/a-goo…
Shivay Lamba retweeted
Today we release mammoth-1 and elephant-1!
We post-trained mammoth-1, the #1 ranked best decision model in the world, and elephant-1, which is the best model under 15B (and beats Jev).
We're open-sourcing all model weights and provide a technical report detailing all our training and data methods.
We teamed up with @huggingface to give away $20,000 to anyone who's building with Beam 🤗
Use Beam for any project - make a demo or a dataset, benchmark the model, create an RL environment - and comment your link in the model's repo page by Monday, November 2.
We choose the winners. Winners choose cash or credits. Claim $25 in Hugging Face credits to build with Beam and enter the competition 👇
huggingface.co/reflection/Be…
I built Kubetrace - using Beam with @nebiustf
xitter.cf/HowDevelop/status/2111…
An AI coding agent becomes useful when it can investigate a problem, change the code, test the fix, and follow through to a working deployment.
That’s what I wanted to explore with the newly launched @reflection_ai Beam 🚀
Beam is built for coding, reasoning and agentic workloads. KubeTrace AI puts those capabilities to work on @kubernetesio incidents, with inference through @nebiustf.
The agent can:
→ Investigate logs, events and application code
→ Trace a failure to its cause
→ Propose code or Kubernetes manifest changes
→ Run regression tests in a sandbox
→ Build and deploy the repaired application
→ Check whether it recovered
The execution layer checks every proposed change against policy, and hidden checks decide whether the incident is resolved. You can inspect the evidence, patches, test output and failed attempts.
Shivay Lamba retweeted
We teamed up with @huggingface to give away $20,000 to anyone who's building with Beam 🤗
Use Beam for any project - make a demo or a dataset, benchmark the model, create an RL environment - and comment your link in the model's repo page by Monday, November 2.
We choose the winners. Winners choose cash or credits. Claim $25 in Hugging Face credits to build with Beam and enter the competition 👇
huggingface.co/reflection/Be…
Shivay Lamba retweeted
Inspired by @BhosalePratim's Macbrow, I built Winbrow: a Windows voice assistant.
@GradiumAI speech + @typesafeai Jev routing + browser automation adapted from @browser_use’s Jev Ultrafast project.
@Alibaba_Qwen 4B LLM handles replies. Phonon (TTS) speaks them both running locally on @Qualcomm_Dev X Elite’s NPU.
Built for Phonon’s OSS launch!
I've always wanted a way to talk to my Mac and do things on the go.
Until now, it was super slow.
So I built Macbrow.
Voice-controlled browser and computer-use agent.
Built using @GradiumAI @browser_use and @typesafeai
Still early and a work in progress.
github.com/timpratim/macbrow
Open source powers software. Software factories can help power open source.
Introducing the SuperPlane OSS Program, free for open-source projects with upto $5,000 in credits for machines and AI models.
Maintainers, apply below 👇
i've joined @nebiusai, the AI-native hyperscaler!
excited to be working with the rockstar devrel team to help devs (and their agents) build with AI at scale.
Shivay Lamba retweeted
Model customization recipes are live on our website and in the cookbook repo.
Beam is now supported across RL and SFT stacks including @axolotl_ai, NeMo RL, NeMo AutoModel, Megatron Bridge (@NVIDIAAI), Miles (@radixark), SkyRL (@NovaSkyAI), and TRL (@huggingface).
Get cooking 🧑🍳
developers.reflection.ai/cus…
The Industrial AI Cloud from @deutschetelekom is the first large-scale industrial AI cloud to put @NVIDIA DSX OS into production, and we automate the private AI cloud each customer gets. Onboarding a new customer used to take days of manual work. Now it takes hours and starts with a click.
Sales engineers, it's done bro! 👋
I'm building whatudo: an AI that does the demo for you. When someone wants to see your product, it hops on a call, shares a real browser, clicks through your product live and answers their questions. No sales team needed for the first demo!
Invite-only for now, link below 👇
This is basically Oscars for Tech Twitter
(and shockingly i am in it hahaha, see cosign.co/awards/rising-star…)
Excited to announce the Cosign Awards: celebrating exceptional people, companies, and products that had a massive impact in 2026.
cosign.co/awards
Anyone can nominate and vote. Award winners will be chosen by the public in conjunction with a panel of judges, including industry leaders like @lennysan, @packyM, @amasad, and many more. We’ll host an event in December honoring the winners (10 per category).
For people categories, we’re mostly focused on everyone except the CEO–everyone else building the company behind the scenes.
Companies with the most amount of people nominations will be listed on our “Top Companies to Work For” page.
True to form, at Product Hunt I helped launch the Golden Kitty awards a decade ago to recognize great products. Soon after I launched the Rise Awards, honoring executives and ICs building companies that make them. The Cosign Awards is the next iteration.
We have 41 categories across people, companies, and products.
People categories:
- Future Founders
- Rising Stars
- Engineering
- FDE
- Product
- Design
- AI Researchers
- Marketing
- Comms
- Growth
- Chief of Staff
- Operations
- Talent
- Finance
- Public Intellectuals
- Podcaster of the Year
- Newsletter Author of the Year
- Poster of the Year
Company categories:
- Rising Stars
- Best Place to Work
- Founder Mafia
- Talent Density
- Highest Aura
Product categories:
- Product of the Year
- Launch of the Year
- AI Assistant
- AI Language Model
- AI Coding
- Dev Tools
- Open Source
- Consumer
- Productivity
- Creative & Media Tools
- Product Design & UX
- Enterprise B2B
- Hardware
- Robotics
- Self-Driving
- Healthcare
- Vertical Fintech
- Legal
🚀 You can now build ExecuTorch on Windows-powered Snapdragon X Elite laptops!
This opens up exciting possibilities for developers building AI applications on accelerated PC hardware.
In this demo, I’m running Llama 3.2 1B locally on my Samsung Galaxy S25 Ultra, showcasing two approaches to on-device inference:
🔹 ExecuTorch: If you have a PyTorch model, you can use ExecuTorch to bring it to on-device inference. The current demo runs on the CPU.
🔹 GenieX: Have a GGUF model? You can run it through GenieX and explore supported inference backends, including NPU , GPU and CPU, depending on your setup.
PyTorch models → ExecuTorch.
GGUF models → GenieX.
Two different paths to experiment with local AI.
🔗 ExecuTorch update: github.com/pytorch/executorc…
🔗 Explore GenieX: aihub.qualcomm.com/geniex
📱 Get started with GenieX: geniex.aihub.qualcomm.com/en…
@PyTorch @UnslothAI @Qualcomm_Dev
An OpenAI agent removed a monitoring integration from code it was migrating.
Asked directly if it had, it said no.
It’s like trusting a student to grade their own homework.
Agent safety monitoring has to live outside the model
That’s exactly why we're building @agenttrailhq
Inspired by @BhosalePratim's Macbrow, I built Winbrow: a Windows voice assistant.
@GradiumAI speech + @typesafeai Jev routing + browser automation adapted from @browser_use’s Jev Ultrafast project.
@Alibaba_Qwen 4B LLM handles replies. Phonon (TTS) speaks them both running locally on @Qualcomm_Dev X Elite’s NPU.
Built for Phonon’s OSS launch!
I've always wanted a way to talk to my Mac and do things on the go.
Until now, it was super slow.
So I built Macbrow.
Voice-controlled browser and computer-use agent.
Built using @GradiumAI @browser_use and @typesafeai
Still early and a work in progress.
github.com/timpratim/macbrow
Fun fact: I've barely ever met @HowDevelop in India 🇮🇳
But outside India? Multiple times. Different countries, different conferences 😄
This time: All Things Open 2026 in Raleigh, with @JuanPa too.
At this point I'm convinced Shivay is everywhere 🌍 #ATO2026
ALT Selfie of Santosh Yadav with Juan Pablo Flores and Shivay Lamba in the hallway at All Things Open 2026, Raleigh
ALT Street banners in Raleigh: 'Welcome to Downtown Raleigh, All Things Open 2026' and 'We're glad you're here!' against a blue sky
@SantoshYadavDev the fun fact actually is despite @HowDevelop lives in BLR, I never met him here. He’s always out of India only and maximum times I met him is out of India 🤣
Shivay Lamba retweeted
I got to build with @GradiumAI Phonon on launch day!
A smart mirror that sees your outfit and gives spoken style recommendations - all running locally on @arduino Ventuno Q.
@Alibaba_Qwen VLM to understand what you’re wearing
Phonon TTS to talk through suggestions
Both accelerated by the @Qualcomm_Dev Hexagon NPU
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
I got to build with @GradiumAI Phonon on launch day!
A smart mirror that sees your outfit and gives spoken style recommendations - all running locally on @arduino Ventuno Q.
@Alibaba_Qwen VLM to understand what you’re wearing
Phonon TTS to talk through suggestions
Both accelerated by the @Qualcomm_Dev Hexagon NPU
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
would love if you could do the same with phonon-wasm! I recently created it to run Phonon with wasm optimizations (using something different from xn-ptts) so would love to see if there's any gain on a regular CPU!