Kartikey ꩜
3.3K posts
Kartikey ꩜
@Kartikey____
Building Hardware 🚶➡️🧘♂️ 🌱
Bangalore, India Katılım Temmuz 2023
90 Takip Edilen114 Takipçiler
According to Indian parents pursuing B.TECh in CSE( Ai/ML) is the solution to every career problem.
English
We implemented @karpathy 's microGPT from scratch in pure C.
Runs at 2.6M tok/s on a Ryzen 5 5600H CPU.
Optimized with AVX2 intrinsics, fixed-width dot product kernels (dot16, dot4, dot64), and Schraudolph's fast exponential approximation.
GitHub: github.com/vixhal-baraiya…
English
English
We implemented @karpathy 's MicroGPT fully on FPGA fabric.
No GPU.
No PyTorch.
No CPU inference loop.
Just a transformer burned into hardware, generating 50,000+ tokens/sec.
The model is small, but the idea is not: inference does not have to live only in software 👇
English
is this a serious ycom post????
reallyy?
Y Combinator@ycombinator
Inference Chips for Agent Workflows @sdianahu Most AI chips are designed for "prompt in, response out." Agents don't work that way. They loop, branch, and hold context across dozens of steps, and current GPUs hit 30–40% utilization as a result. That gap is where purpose-built silicon wins.
English
Bangalore Job seekers Kit
Sheet link- #gid=116538532" target="_blank" rel="nofollow noopener">docs.google.com/spreadsheets/d…
Startup Bundle- topmate.io/letscode/20711… .
English
Waiting
Y Combinator@ycombinator
Inference Chips for Agent Workflows @sdianahu Most AI chips are designed for "prompt in, response out." Agents don't work that way. They loop, branch, and hold context across dozens of steps, and current GPUs hit 30–40% utilization as a result. That gap is where purpose-built silicon wins.
English
Kartikey ꩜ retweetledi
@DanielleFong @dwarkesh_sp It's built link b/w classroom to workplace.
English
dwarkesh continuing to reinvent the university
Dwarkesh Patel@dwarkesh_sp
Did a very different format with @reinerpope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. – How batch size affects token cost and speed – How MoE models are laid out across GPU racks – How pipeline parallelism spreads model layers across racks – Why Ilya said, “As we now know, pipelining is not wise.” – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal – Deducing long context memory costs from API pricing – Convergent evolution between neural nets and cryptography
English
I built an IDE that writes your hardware code.
Describe what you want to build,
get a wiring diagram, the code, and one-click upload to 720+ boards!
Completely free. BYOK if you want.
ide.blueprint.am
English
Announcing @slashyai x @attio.
The first email client to have a native Attio integration.
Your full Attio context now lives in your inbox -- right where you actually move deals forward.
Everything syncs instantly between Slashy and Attio.
You can even update Attio straight from iMessage or Slack.
Finally there's an AI-native inbox, for the AI-native CRM 🤝
English
Kartikey ꩜ retweetledi
49 drones. One pulse. All gone.
Beyond the tech, it shows a shift where scale and economics matter as much as firepower.
Recent conflicts highlight a brutal reality: cheap kamikaze drones cost a fraction of the interceptors sent to destroy them.
The aggressor doesn’t need to win. He just needs to keep the math working in his favour.
And while lasers are much cheaper & great for precision, they only engage one target at a time. Against a swarm, that’s a problem.
HPM doesn’t have that constraint. It covers a volume of space, not a point.
Both are meant to complement kinetic systems (missiles, guns) rather than replace them. The future of air defense is clearly layered, with each technology filling a different niche.
For India, this is very pertinent. Importing solutions reactively isn't a strategy.
Building indigenous, AI-enabled HPM and laser capability early is.
We have the talent. We just need faster procurement, patient capital, and institutions that let deep-tech startups scale.
On a personal note, I’ve recently taken on the role of Chairman of iCreate, a leading deep-tech incubator in Gujarat.
I would like it to be the home for exactly this kind of innovation.
If you’re building the technologies that will define tomorrow’s defense, do check it out at icreate.org.in
English
This is interesting. 🙌
Dave Jones@eevblog
May 5th will hence forth be known as International 555 Timer Day. May 5th at 5:55 to be precise.
English
Working at BEL BGCX and just got drafted for Karnataka Census duty. While I joined to build tech, I'm now on the ground for a national cause.
Bharat Electronics Limited (BEL)@BEL_CorpCom
A virtual online outreach session is being organised by BEL in association with iDEX-DIO/DDP-MoD, to disseminate information on BEL’s DRISHTI (DPSU-driven Research & Innovation for Strategic and High-impact Technology Integration) Challenges / Problem Statements. @DefProdnIndia
English
A Peenya equivalent of Shenzhen is the only way to stay in the game.
Kyle Chan@kyleichan
Robotics firms in Shenzhen by segment: - Control systems: ~3,600 - Cloud “brain”: ~3,000 - Energy & communications systems: ~1,700 - Actuation systems: ~600 - Sensing/perception systems: ~550 - Drive systems: ~420 - Structural systems: ~330 - Transmission systems: ~160 From the Shenzhen Robot Industry Development White Paper (2025) 深圳市机器人产业发展白皮书（2025年） m.thepaper.cn/newsDetail_for…
English