Infrastructure is going autonomous. The spending has to stay accountable.
Vyomedha builds intelligence for cloud and AI infrastructure — one view of spend, the waste surfaced, performance protected across every provider you run. Pellid, our first product, is in private development: an advisor today, built to earn its way to acting safely on your behalf.
Most tools tell you what's wrong. We're building the one that acts.
Every cloud and AI bill hides waste. The industry's answer has been a dashboard of red numbers — and the fixing left to you. Vyomedha takes the next step: systems that understand your entire cloud and AI footprint, find the waste, and hand you the exact change to make — why, how much you'll save, the risk, and how to undo it.
Cloud + AI cost in ONE view
Almost nobody unifies COGS across both. We bring cloud and AI spend into a single, coherent telemetry stream.
AI waste as a real FinOps loop
Detect → recommend with risk tier + undo → prove rupees saved. Not just dashboards, but action.
INR / GST-native billing
AWS Marketplace billing and GST input credit modeling. Global tools are USD-irrational in India.
Each wave represents a model's live request volume over time. The calm mint band indicates healthy traffic; a red spike signifies waste Pellid catches in real-time — a retry storm, a context-bloat loop, or a job routed to an unnecessarily expensive model. Watch each spike flatten back to baseline: that's Pellid intervening before the invoice generates.
0 this session
0 this session
0 this session
Pellid monitors the AI bill identically to the cloud bill — routing each task to the cheapest model that preserves quality, catching retry storms, and severing token waste before it materializes on the invoice. Signals shown are illustrative.
We earn the right to act by proving value first — read-only until you say otherwise.
We hand over a reviewable change; the customer applies it. Safety is a feature.
"They saved us real money, and we could prove it" is the sentence we're building toward.
Projected vs realized savings tracked to train a recommendation engine on real Indian infrastructure.
See what we're building.
A 2:15 story film about the problem Pellid is built to solve.
A continuous, ambient loop.
Observing your infrastructure quietly.
Idle
Observing cloud & AI, quietly — no noise until something matters.
Scanning
Parsing every resource and cost signal across both bills.
Analyzing
Isolating actual waste from required spend, defining the safer path.
Deploying
Surfacing the exact fix — ₹ saved, risk tier, and the undo.
Complete
Upon approval, savings reconcile against the real invoice.
The waste, classified.
Cloud · AWS / Azure
- Idle compute & zombie instances
- Orphaned storage & unattached disks
- Over-provisioning & right-sizing gaps
- Unused reservations & commitment gaps
- NAT gateway & data-transfer leaks
AI · Claude / GPT / Sarvam
- Model routing inefficiencies
- Missing caching & prompt dedup
- Retry storms & cascading failures
- Token waste & context bloat
- Cost per outcome vs cost per token
We never touch production until we've earned it.
Read-only access first. Always.
Trust is the price of action, and we pay it rupee by rupee.
Every recommendation is reversible.
Safety is a feature, not a limitation. If an action can't be undone, we won't recommend it.
Proven with an audit trail.
Every rupee of savings will be proven with an audit trail — deep links to the source bill, not estimates.
We recommend, you approve.
Always. We hand over a reviewable change; the customer applies it.
Optimization is how we get in. Becoming the operating layer of the cloud is where we go.
Show what you spend, across cloud and AI, in one place.
Show where the waste is, and how much can be recovered.
Tell you exactly what to change: why, rupees saved, risk, how to undo.
Generate the change (IaC pull request / prompt) for you to review and apply.
Describe the outcome you want, and the system provisions it safely. The DevOps and FinOps team in one.
Visibility isn't the answer; action is. But action without trust is dangerous.
So we build systems that earn trust slowly, prove every result, and only then ask for more access. More infrastructure intelligence systems are coming. Vyomedha is a house, not a single product. The architecture supports the full compute infrastructure lifecycle.
Trust before features
We earn the right to act by proving value first — read-only until you say otherwise.
Recommend, don't touch
We hand over a reviewable change; the customer applies it. Safety is a feature.
One by one
Build, get feedback, improve, repeat. No premature complexity. Small team, long horizon.
Two builders, one conviction.
Engineers who got tired of tools that point at problems and walk away. Incubated at iHub, Ahmedabad, building in the careful, one-step-at-a-time way infrastructure demands: prove it, earn trust, then earn more. Small team, long horizon.
Harsh Patel
Co-founder & CEOHarsh brings direct operational experience with enterprise infrastructure and resilience systems. He has seen firsthand how infrastructure management fails when dependency maps are wrong or stale — the experience that shapes how Vyomedha builds infrastructure intelligence that stays accountable.
Maulik Kansoda
Co-founder & CTOMaulik leads the engineering and productization track — building Vyomedha into a real multi-tenant SaaS platform with authentication, RBAC, team management, and cloud connectors.
We're building this with a few teams, not for them.
Pellid is in private development. We're onboarding a small group of design partners — teams running real cloud and AI spend who want to shape the tool to their architecture.
Direct founder access
Weekly check-ins with Harsh and Maulik. Real conversations, not support tickets. Your feedback goes straight into the product.
Founding-era terms
Design partners get founding-era terms — we'll honour the partnership when we set pricing.
Shape the roadmap
Tell us what matters to your infrastructure. Your priorities move up the queue. This is your chance to build the tool you actually need.
Applications open now · A small first group
Apply as a design partner