
Founder, @CumulusLabsIO @YCombinator (W26)
San Francisco, California
Joined January 2026
- Tweets67
- Following70
- Followers127
- Likes113
Veer Shah retweeted
We've raised $4.9M to build @tndminc - The AI system for hardware engineering.
Engineering ambition has outgrown the systems built to carry it. Tandem connects requirements, design intent, CAD changes, reviews, and validation.
The next great machine is waiting on its system.
It’s official: @HridayUnadkat and I are launching @tryripline, where we’re automating work for critical industries.
The problem we’re solving is personal.
At @SpaceX, I saw the edge up close: operational efficiency. Most hardware companies never get there.
And after working on simulations and starship at @SpaceX, Anwith and I kept seeing the same problem: hardware teams are being asked to move faster, while the operations behind production are getting more complex.
That’s why we’re building Ripline: we deploy agents to do the operational work that slows hardware teams down.
If you’re building hardware and these stories sound painfully familiar, I’d love to talk. DM me or check my calendly link below.
Veer Shah retweeted
We're on track to hit $4M ARR in under a year. 🚀
Getting from $0 to $20k MRR was by far the hardest part.
So we documented everything that got us there:
• Every traffic channel we used
• Our daily growth playbook
• The exact outreach scripts and messages
• The systems we repeated every single day
No fluff. Just the playbook that got us our first $20k MRR.
Want it?
Repost ♻️ so more founders can see it.
Comment "20k" and I'll DM it to you.
dropped the blueprint on model training at @CumulusLabsIO – sft to online rl. see if you're actually building shit: cumulus.blog/visual-generati…
2
7
263
available models:
xitter.cf/VeerCumulus/status/202…
Replying to @VeerCumulus @CumulusLabsIO
we added a ton of models, lmk if you want any more but we have:
- VLMs, LLMs (Qwen3.5, KimiK2.5, more)
- video gen
- image gen
- audio gen
- tts
1
2
124
just released ionrouter.io @CumulusLabsIO
we switched our OpenCode & OpenClaw setups to use it; KimiK2.5 is blazing fast— way better than openrouter
3
5
207
Veer Shah retweeted
We just launched IonRouter: ionrouter.io
One API for the best open‑source models — LLMs, vision models, TTS, and video generation.
Veer Shah retweeted
Host anything. Bring your own LoRAs, finetunes, or any open‑source model — we’ll serve it on our fleet.
No GPU setup, no infra management, no idle cost.
Pay nearly half of what you would with other providers.
they say cambridge is the smartest sq. mi in the world but what about impactful? @CumulusLabsIO @ycombinator
CS majors: We built a custom stack on GH200s that CRUSHES Together AI's throughput on 4 parallel vLLMs (details + benchmarks: cumulus.blog/ionattention)
Beat our numbers → instant interview at a YC Startup.
DM resume + drop a comment if you've already figured out how we did it 😈
Tag a friend who thinks they can beat us
2
9
310
Veer Shah retweeted
Grace Hopper doesn’t feel like “an H100 with more RAM” in practice. It feels like a weird little supercomputer in one box. Once we stopped treating it like a normal GPU and built around it properly, our 7B models suddenly started doing thousands of tokens/sec on a single chip.
Challenge to all CS majors: we just made an inference stack on GH200s thats got higher throughput than TogetherAI on 4 vLLMs (peep cumulus.blog). If you can beat us, you get an interview, dm us your resume and comment if you've figured it out.
5
122
Veer Shah retweeted
OctaPulse (@tryoctapulse) uses robotics and computer vision to automate fish inspection for aquaculture farms. From hatchery to processing, they're building the future of autonomous aquaculture.
Congrats on the launch, Paul and @rohxnsxngh!
ycombinator.com/launches/PUE…
If you're paying for inference APIs, Cumulus lets you run the same models yourself for less. If you're self-hosting, Cumulus makes your models faster without the operational burden. If you need GPUs for any workload at all, Cumulus is the simplest and most performant way to get them.
cumuluslabs.io
Modal makes you wait. Together makes you pay for always-on.
Cumulus gives you both — serverless GPUs + a real inference stack (ION) that outperforms stock vLLM/SGLang with zero setup.
Pay by the second. Scale to 0. Cold starts under 10s.
Let's talk → calendar.app.google/DfDdq1fN…













