AI support agents · Built in India
Most support tickets are the same handful of questions arriving again and again. I build AI agents that answer those automatically — so your team spends its day on the ones that genuinely need a person.
The real cost
Your support team isn't the problem — the fact that they answer the same question forty times a week is.
How do I reset this, where's my invoice, why won't it print. Skilled people spending their day as a search engine.
Customers in other timezones, or a clinic opening at 7am. The queue grows while everyone sleeps.
Front desks and ops teams churn. The same onboarding questions come back with each new person, not once at go-live.
What it actually does
Rule-based tools break the moment a customer phrases something differently. These agents reason about the question, then either answer it or hand it over properly.
01
Your FAQs, help articles, product docs and past support threads. It answers the way your team already answers, in your language and your terms.
02
Checks an order, pulls an invoice, looks up a booking, resets access. It does the thing rather than describing how to do it.
03
When it shouldn't answer, it doesn't guess. Your team receives the ticket with the history, the account and what's already been tried.
Proof
Every one of these is deployed and clickable, with the source public. Open them before you reply to me.
A support widget for a salon and spa platform — answers product questions from a knowledge base, escalates to a human with context, and drops into an existing site. Built to demonstrate the exact category of thing I'd build for you, not for a client.
Ask questions across hours of a YouTube channel's content and get accurate, sourced answers. This is the same retrieval architecture that lets a support agent answer from your documentation instead of inventing things.
Books flights, files complaints and looks up policy end to end. The clearest demonstration that these agents complete tasks rather than just replying — which is what separates support automation from a chat window.
Speak a question, get a spoken answer, with live search underneath. If your support arrives by phone rather than email, this is the piece that matters.
Several agents researching in parallel and assembling a finished article. Included because it shows the orchestration that keeps a support agent reliable once it has more than one job to do.
How it goes
You'll know where it stands every day. No retainer to sign before you see it working.
Day 1
Help docs, FAQs, a export of past tickets — whatever exists. If your documentation is a mess, that's normal and it's still enough.
Days 2–6
Retrieval, memory, the integrations into your helpdesk or WhatsApp. You get progress daily and can redirect it while it's cheap to change.
Day 7
Deployed, with the full source, documentation and a handoff call. If we never speak again it keeps running.
Pricing
A fixed build fee, not a per-seat subscription that grows as you do. Ongoing updates are optional and separate.
Starter
one-time · live in 5 days
Support
one-time · live in 7 days
System
one-time · live in 14 days
The build fee is one-time and the code is yours. Hosting and model usage are billed by the providers directly, typically ₹500–2,000 a month depending on volume. A monthly retainer is available if you'd rather I keep improving it, but nothing requires one.
Before you ask
No. You'd be the first, and you should price that in — which is exactly why it costs ₹40–90,000 rather than what an agency would charge for the same build. Here's what I think you actually get from being early: my full attention rather than a queue position, and a founder who needs this to work more than you do. Here's what protects you: the code is yours from day seven, so if I turn out to be wrong about any of it, you're holding a working system rather than a cancelled subscription. Five agents above are live and open source — judge the work, not the client list.
It will, if nobody updates what it knows — that's true of your help docs too. Two things stop it being a problem. The agent answers from your documentation rather than from memorised rules, so updating a help article updates its answers without touching any code. And when it hits something outside what it knows, it escalates instead of guessing, so a stale agent degrades into a slower agent rather than a wrong one. If you'd rather I keep tuning it as you ship, that's what the monthly retainer is for — but it's a convenience, not a dependency.
It says so and hands the ticket to your team rather than guessing. This is a design decision, not a limitation — an agent that invents an answer about billing or compliance costs you more than one that escalates. You set where that line sits, and you can watch what it escalates and tighten it over time.
They could. The question is whether you want them to. Support tooling competes with your actual product roadmap, and it's the kind of build that looks like two weeks and turns into two months of edge cases. I've built five of these end to end, so you're buying the version that already went through those mistakes.
It's trained on how your team already writes, so it picks up your phrasing rather than generic support language. And you review it before it faces a single customer — nothing goes live until you've read what it says.
You keep everything. The source code, the documentation, the deployment. That's the reason it's a one-time build rather than a subscription — you shouldn't be locked into me to keep your own support running.
Wali — I study CS and Applied Maths at IIIT Delhi and build AI agents. That's the whole answer: no agency, no account manager, no team of subcontractors. You talk to the person writing the code, and the five agents above are all mine and all open source, so you can judge the work before you commit to anything.
Tell me roughly what your support queue looks like and I'll write back with what I'd actually build for you — which questions get answered automatically, what still reaches your team, and where it would sit on top of whatever you use now. No cost, no call, no deck. If I don't think it's worth building, I'll tell you that instead.
Got it — I've got what I need, I'll write back soon.