About
We build agents that get measured against a real number.
Noesa is the trading name of Finnodiagnos Analytics PVT Ltd, based in Bengaluru. We build and run AI agents for Indian businesses — and we are fairly particular about which problems deserve one.
What we actually do
Most of our work is one shape: a job inside a business that is repetitive, conversational and currently eating a person’s day. Order status. Appointment booking. Chasing an incomplete document file. Qualifying an inbound lead at 11pm so it is still warm at 10am. We build an agent that does that job on WhatsApp, connect it to the systems that hold the real answers, and then keep running it — because an agent that nobody watches drifts away from the business it was built for.
Around that sits the rest of it: automation, data engineering, and the software that has to exist before an agent has anything true to say.
How we scope work
- It starts with a number you already track. Not a number we invent for the proposal — one that already exists in your business and that someone already cares about. Messages answered without a human. Applications that reach a complete file. Renewals completed before the lapse date. If we cannot find one, that is the finding.
- The escalation path is designed first. What the agent must never decide is a more important specification than what it can do. In lending it is the credit decision; in insurance it is claim payability; in a clinic it is anything clinical. Those route to a human by design, not by failure.
- You get the mechanism, not a black box. How the agent is grounded, where its answers come from, what happens when the source is silent. If a vendor cannot explain what their system does when it does not know something, that is the answer to your question.
What this site deliberately does not do
The guides and comparisons follow rules we hold ourselves to, and they are worth stating plainly because they explain some of the gaps you will notice:
- No invented numbers. Where a figure moves often — per-message rates, subscription prices — we publish a band and point at the source rather than quoting a precise value that will be wrong within a quarter.
- Competitor claims are dated. Every comparison page says which day it was checked against that vendor’s own live site. One of them had to be rewritten mid-draft because the competitor had repositioned since we planned it.
- We say when we could not verify something. Several guides carry an explicit note where Meta’s own documentation is missing, contradictory or returns a 404. Writing around those gaps confidently would be easier and worse.
- We recommend competitors where they fit. If what you need is a staffed shared inbox rather than an autonomous agent, the comparison pages say so outright and name who does it well.
Talk to us
The fastest way to find out whether this is worth your time is to message the product and then message a human — both numbers are on the contact page. One sentence about the job eating your team’s day is enough to get a straight answer, including the answer that you do not need us.
Frequently asked questions
- Who is Noesa?
- Noesa is the trading name of Finnodiagnos Analytics PVT Ltd, a company based in Bengaluru, India. We build and run AI agents for businesses — mostly on WhatsApp, because that is where Indian customers already are. Vaani is our own WhatsApp agent product, and it is live on a public number rather than behind a demo request form.
- Why are there no case studies or client logos on this site?
- Because we are not going to invent them, and real ones need a client's permission that we would rather ask for properly than assume. That is a genuine gap and we would rather you notice it than have it papered over with stock photos and unverifiable percentages. What we offer instead is a live product you can test yourself in one message, and pages that explain the mechanism rather than asserting a result. When there are case studies worth publishing, they will name the client and the number.
- What kind of work do you turn down?
- Anything without a number attached to it. If nobody can say what would be different in six weeks, the project has no definition of working and no way to end. We also turn down agent projects that are really automation projects — a deterministic workflow is cheaper, more reliable and fails more loudly than an agent, and quite a lot of what gets sold as AI should just be a rule. We wrote the whole argument down, including the part that argues against our own pipeline.
- How do you decide whether something should be an AI agent at all?
- The test is failure asymmetry. Automation fails loudly and within bounds — it stops, and someone sees it stop. An agent fails quietly and plausibly: it gives an answer that reads fine and is wrong. If the cost of a quiet, plausible mistake in your process is high, the agent needs hard limits and a designed escalation path, or it should not be an agent. That is a design decision made before anything is built, not a setting adjusted afterwards.
- What does the name mean?
- Noesa comes from noesis — the Greek for understanding or direct apprehension, as distinct from mere calculation. It is a deliberate distinction: the difference between a system that matches your words against a script and one that works out what you actually meant. Vaani is Sanskrit for speech or voice, which is what the product does with that understanding.