Skip to content
UpgradeTechno

AI engineering studio — Chennai

Production AI systems for Indian SMBs.

We build the retrieval and agent systems the large firms will not quote under ₹20 lakh, at a pace and price that fits a business your size.

Book a 15-min callSee a live demo
  • Fixed scope, fixed price
  • First version live in 10 days
  • You own the code and the data

The gap

Why most SMBs stall before anything ships

It is rarely the technology. It is that the two kinds of vendor available to a mid-sized Indian business both fail in predictable ways.

  • The good firms will not quote you

    Established AI consultancies here run on team-month economics. Below roughly ₹20 lakh a project does not clear their internal bar, so a ₹4 lakh problem never gets a serious proposal — and you are left assuming the work is out of reach.

  • Automation agencies hit a ceiling

    A no-code flow gets you a convincing demo in a fortnight. Then real customers arrive with vague questions, half-typed phone numbers and edge cases, and there is no way to add retrieval, evaluation or a fallback path. The demo does not become a system.

  • Pilots that never get handed over

    Plenty of projects reach a working prototype and stop there — no deployment story, no monitoring, no one who can change a prompt after the invoice clears. A pilot nobody owns is indistinguishable from no pilot at all.

The offer

AI Front Desk

A WhatsApp front desk for clinics and diagnostic labs

One productised system, built the same careful way each time. It answers patient queries in the language they write in, books and reschedules appointments against your existing calendar, chases no-shows before the slot is wasted, and delivers reports when they are ready.

  • WhatsApp Business API, properly set up

    Verified sender, message templates submitted and approved, opt-in handling that matches Meta policy. Patients message the number you already advertise.

  • Answers grounded in your actual documents

    Timings, preparation instructions, test panels, price lists and doctor availability are indexed with retrieval, so the assistant quotes your material instead of improvising.

  • Booking, rescheduling and cancellation

    Two-way sync with the calendar or HMIS you already use. Slot collisions, buffer times and doctor leave are handled in the booking logic, not left to the model.

  • No-show and follow-up sequences

    Reminders on your schedule, a confirm-or-release step before the appointment, and a re-book prompt for patients who drop off.

  • Report delivery

    Reports pushed to the patient on WhatsApp once released, with identity checks before anything is sent and a full record of what went to whom.

  • Escalation to a human

    Anything clinical, ambiguous or upset is routed to your front-desk staff with the conversation history attached. The assistant is explicitly not allowed to give medical advice.

Ten working days, start to switch-on

  1. Days 0–2

    Discovery

    We sit with your front desk for a morning, read the last few hundred real patient messages, and write down every question they actually ask.

  2. Days 3–6

    Build

    Retrieval over your documents, booking logic against your calendar, WhatsApp templates submitted for approval while the rest is being built.

  3. Days 7–8

    Evaluation

    We run the assistant against the real question set from day one and fix what it gets wrong. You see the pass rate before anything goes live.

  4. Day 9

    Supervised pilot

    Live with a subset of patients, every conversation reviewed by your staff, escalation path tested under real load.

  5. Day 10

    Handover

    Full switch-on, a walkthrough with your team, and written runbooks for the things you will want to change yourself.

Fixed price, quoted in full on the first call, payable in two instalments. No hourly billing and no change-request meter running in the background.

Full scope and boundaries

How we work

Three phases, and you get something real at the end of each

No phase ends with a status update. Each one produces an artefact you keep whether or not you continue to the next.

  1. Discovery

    Two days understanding the work before proposing to automate any of it. We read your real message history rather than guessing at it.

    What you get

    • A written scope with what is explicitly out of it
    • The evaluation set — the real questions we will grade against
    • A fixed price and a dated delivery plan

  2. Build

    One engineer, working in the open. You get a staging link from day three and can watch it improve rather than waiting for a reveal.

    What you get

    • A staging environment you can message from your own phone
    • Weekly evaluation scores against the day-one question set
    • Source code in your GitHub organisation from the first commit

  3. Handover

    The phase most projects skip. We are finished when your team can operate and change the system without us.

    What you get

    • Runbooks for the changes you will make most often
    • A recorded walkthrough with your front-desk staff
    • Thirty days of support included, then a plain monthly rate or nothing at all

Also build

Full-stack product engineering

The AI work sits on top of ordinary, well-built software. When a client needs the surrounding product too — a dashboard, an internal tool, an API — we build that as well.

Front end

  • React
  • Next.js
  • TypeScript

Back end

  • Node.js
  • Python
  • FastAPI
  • GraphQL

Data

  • PostgreSQL
  • MongoDB
  • Vector search

AI

  • RAG pipelines
  • Agentic workflows
  • Evaluation harnesses

Questions

The six things everyone asks

What does it cost?
AI Front Desk is a fixed price, quoted in full on the first call once we know how many locations and doctors are involved, and payable in two instalments. There is no hourly billing. Running costs — WhatsApp conversation charges and model usage — are billed to your own accounts, so you see them directly and we have no incentive to inflate them.
Is ten days realistic, or is that a marketing number?
Ten working days from the discovery session to switch-on, and it is realistic because the scope is fixed. We are not designing a new product each time — we are configuring a system we have already built against your documents, your calendar and your question set. The one thing outside our control is WhatsApp template approval from Meta, which we submit on day three specifically so it runs in parallel.
What happens to patient data?
It stays in your infrastructure and your accounts. We deploy into a cloud project you own, we do not copy conversation logs or patient records into anything of ours, and we sign whatever data agreement your compliance team requires before discovery starts. Message content is not used to train any model. Access we are given for the build is revoked at handover unless you keep us on support.
What stops it from saying something wrong to a patient?
Three things. Answers are grounded in your documents through retrieval rather than generated from memory, so it quotes your timings and your preparation instructions. It is instructed and tested to refuse anything clinical and hand off to your staff instead. And before go-live we grade it against a set of real questions taken from your own message history — you see that pass rate and decide whether it is good enough.
What happens after handover? Am I locked in?
No. The source code sits in your GitHub organisation from the first commit and the system runs in your cloud account, so you can hand it to any competent engineer. Thirty days of support are included after switch-on. After that you can take a plain monthly support rate, call us only when you need something, or not call us at all — the system keeps running either way.
We are not a clinic. Will this work for us?
The productised ten-day version is built specifically for clinics and diagnostic labs, and that is where it is worth the fixed price. The underlying work — retrieval over your documents, agents that take real actions, an evaluation harness that proves it works — is not sector-specific. If you have a high-volume repetitive enquiry problem somewhere else, say so on the call and we will tell you honestly whether it is a two-week project or a two-month one.

Tell us what your front desk repeats all day.

Fifteen minutes is enough to say whether this is worth building for you. If it is not, we will tell you that on the call rather than sell you a discovery phase.

Book a 15-min callEmail hello@upgradetechno.com