Dhansurrya S P Start a conversation

AI Product Engineer · Chennai

Shipped.Not demoed.

I'm Dhansurrya. I build AI that works on real websites, for real people. My AI assistant is live on four brand websites today.1

Start a conversation See the work
Dhansurrya S P, arms crossed, wearing glasses and a black t-shirt against a yellow wall

I turn messy knowledge into clear answers.

Grounded in your content. Fast. Checked before it ships.

Prototype. It answers only from this page's text and calls no AI model.

Hi again. I'm Dhansurrya.

Open to full-time roles (Chennai or Bengaluru) and freelance projects.

Download resume

I'm an AI product engineer in Chennai. I spent four years building websites for agencies and global brands, so I care about the parts people actually touch. Now I build AI features that hold up on real sites: grounded answers1, fast streaming2, and code I've checked myself3.

  1. 1

    Every answer comes from that site's own content, not a guess.

    Multi-brand AI assistant
  2. 2

    Answers stream in live, word by word.

    Multi-brand AI assistant
  3. 3

    I use AI tools to move faster, then I read, test and fix every line before it ships.

    Fair questions

Four years of shipping

  1. Mar 2026 to now

    Senior Web Developer · HealthBound Digital

    Lead frontend and AI product work. Built and shipped the multi-brand AI assistant.

  2. Oct 2025 to Mar 2026

    Career break

  3. Jul 2024 to Oct 2025

    Front End Developer · Havas Worldwide (Prose on Pixels)

    Campaign sites and brand pages for global brands, built from reusable parts.

  4. Jul 2022 to Jul 2024

    Web Developer · HTML Kitchen

    20+ production pages and custom WordPress themes for clients.

In production

Easy to demo. Hard to trust.

So I build the trust part. This assistant runs on four brand websites, from one codebase.

One build, four brands

Visual recreation with sample data. The real product UI is confidential.

Company project, developed as part of my professional work.

Multi-brand AI assistant

Employer project. Client names kept private.

Problem
Easy to demo. Hard to trust.
My role
Built and shipped the multi-brand AI assistant.
Stack
React · TypeScript · SSE streaming · RAG · Claude API · Gemini · Supabase · Sentry

How it works

  1. AskA visitor asks a question on a brand site.
  2. ProtectAPI keys stay on the server, with rate limits and bot protection.
  3. GroundThe answer comes from that site's own content.
  4. GenerateClaude API, Gemini.
  5. StreamWords stream in live, with the source.
  6. WatchError tracking and daily reports.

Simplified from the project notes. The real system is confidential.

What it does
  1. 1Answers stream in live, word by word.
  2. 2Every answer comes from that site's own content, not a guess.
  3. 3One codebase, four brands, each with its own colors and setup.
  4. 4API keys stay on the server, with rate limits and bot protection.
  5. 5Works with a keyboard, screen readers and reduced motion.
  6. 6Watched in production with error tracking and daily reports.

Personal project

tinygpt

A small GPT-style language model, built from scratch in PyTorch with no proprietary AI APIs.

Python · PyTorch

  1. Datatokenizer
  2. Modelself-attention
  3. Trainingbackprop
  4. Outputtext generation

Steps from the project README, grouped into four stages.

Website concepts

Fictional websites I designed and built to practise art direction. None of them is a real business.

Halden Studio desktop homepage concept

An editorial site for an architecture studio, with a drawn axonometric instead of photography.

Architecture practiceConcept projectView concept

More things I've built

  • Contact data platform

    Company project, developed as part of my professional work.

    Search, filter and import hundreds of thousands of records. Tested in CI.

    React · TypeScript · Vitest · NestJS

  • React analytics dashboard

    Personal project

    Charts, themes and reusable parts, built for fun and practice.

    React · Vite · Recharts

How I build

Three steps. No shortcuts.

  1. 01

    Ground it

    Answers come from your real content. If it isn't in the source, the AI shouldn't say it.

  2. 02

    Ship it

    Streaming, testing, safe keys, real websites. Not a notebook, not a slide.

  3. 03

    Watch it

    Errors tracked, conversations measured, reports every day. I know when something breaks.

Do you ship to the EU?

Probably, most stores do.

Yes, to all EU countries.Source: Shipping policy, section 2

Press and hold to ground the answer.

That's the job. Every time.

Toolkit

AI Engineering
Claude API, Gemini, RAG, prompt design, streaming chat
Frontend Engineering
React, TypeScript, Tailwind, Framer Motion
CMS & Web Development
WordPress
Backend & APIs
Node.js, Supabase
Developer Tools
Vitest, Testing Library, GitHub Actions, Sentry, Vercel, Netlify

Fair questions

Can you take an AI feature from idea to production?

Yes. My assistant went from first prototype to four live brand websites, with tests, monitoring and safe API keys.

How do you stop the AI from making things up?

I ground every answer in your own content and keep the model to what the source says. Then I test it on real questions before it goes live.

Do you just paste what AI tools give you?

No. I use AI tools to move faster, then I read, test and fix every line before it ships.

Do you take freelance projects?

Yes. Tell me what you want to build and I'll tell you honestly if I'm the right fit.

Where are you based?

Chennai, India. Open to moving to Bengaluru, and happy to work remotely.

Contact 06

Let's build what's next.

Have an idea, a product, or a problem to solve? I'm always up for a conversation around AI products, web development, or healthcare technology.

Email

dhansurrya2001@gmail.com
This is about

Opens your email app with this message filled in. Nothing is sent from this page.