Tvashta Labs

Tvashta Labs builds lightweight GPU orchestration for datacenters and a unified platform for ML engineers to develop and inference AI models, all without the complexity of traditional infrastructure.

Get in Touch

What We Build

We turn raw datacenter hardware into a platform with all AI tools integrated, ready for ML engineers to use without any setup.

MeghCloud

Unified single plane platform for ML Engineers to develop and inference AI Models.

High Performance Tools like Vajra

Cost-saving and performance-maximizing tools for GPU datacenters. Includes our fast AI model weights streamer.

Niyanta

Lean, bare-metal orchestration for GPU datacenters.

Bare-metal GPUs
Datacenter hardware

Vajra, Our Model Streamer

Vajra, our AI model weights streamer, beats NVIDIA Run:ai and Hugging Face hf_transfer in model load times.

Model load time (Lower is better)
Vajra
8.22s
NVIDIA Run:ai
15.85s
Hugging Face hf_transfer
36.88s
Performance delta
350% (4.5× faster)
Try Vajra

How It Works

Without Tvashta Labs
ML Engineers get GPUs from platform A, use tool B and tool C, and stitch together multiple workflows.
Setup is unnecessarily complex. Engineers need to learn about unrelated infrastructure just to start training.
Datacenters have raw GPUs, but orchestrating them for AI workloads is incredibly difficult.
Existing orchestration software (like OpenStack) is too heavy, bloated, and slow for modern AI needs.
With Tvashta Labs
One platform with all AI tools integrated, ready for ML engineers to use without any setup.
What used to take days of engineering time to set up, now can be set up in a click.
Lightweight orchestration that manages bare-metal GPUs, faster and leaner than OpenStack.
Directly increases utilization, revenue, and contract value for GPU datacenters.

Tvashta Labs is designed for Tier 2 and Tier 3 regional GPU datacenters, and enterprises building on-premise private AI clouds.

Team

Ishan Khare

  • One of the first engineers at vCluster (helped build the core product and scale from Seed to Series A).
  • Deep expertise in datacenter virtualization; published a book on KubeVirt based on work with a GPU Datacenter.
  • Speaker at KubeCon EU Valencia 2022.
LinkedIn

Shivang K Raghuvanshi

  • Specializes in systems programming and container networking.
  • Worked on Podman's core network management tools in Rust during the Linux Foundation's LFX Mentorship.
  • Speaker at KubeCon EU Amsterdam 2026.
LinkedIn

Let's build the future of AI infrastructure together.

Ishan Khare · ishan@tvashta-labs.in

Shivang K Raghuvanshi · shivkr@tvashta-labs.in