Guruprasad Lakshmanan

Independent software and AI consultant, Chennai, India

Hello, I'm Guru. For twenty-five years I have built software that people rely on, first writing it myself, then leading a 150-person engineering division. Since 2021 I have worked independently, helping businesses put AI to work in ways that are useful, reliable and easy to check.

Get in touch

A kolam is one unbroken line drawn around a grid of dots, a morning tradition across Tamil Nadu. It is also quiet mathematics: a few simple rules, followed carefully, give a pattern that never breaks. That is how I like to build software. This one is generated fresh for your visit.

Selected work

A few things I have built as an independent consultant. My clients trust me with their business, so I don't name them; here is what the work does, and what it runs on.

  1. Outpatient payment grouping

    A rules engine that sorts outpatient hospital visits into the groups used for payment, replacing coding by hand. The regulator's manuals were turned into data, so a rule change means re-running, not re-writing.

    Checked against the regulator's own worked examples, and found errors in the official rulebook.

    Python with no outside dependencies, automated browser checks

  2. Market intelligence workspace

    Scores opportunities across product lines and countries, keeps watch on competitors, and answers questions in plain language. The AI model only chooses which tools to use; ordinary code calculates every number.

    Every figure can be traced back and checked.

    LangGraph agent, text-to-SQL, DuckDB, Python

  3. AI product photography

    Five image models working together to turn plain product shots into catalogue photographs, with models, backgrounds and accessories generated to order.

    Cost per product down from over USD 100 to USD 2-3.

    Diffusion and vision models, FastAPI, credit-based billing

  4. Virtual try-on app

    Shoppers see garments on their own photos before they buy.

    Live on iOS and Android.

    Flutter, FastAPI, Gemini vision

  5. Ordering over WhatsApp

    Customers order by chatting in English, Tamil or Tanglish. An AI agent with nine tools matches what they ask for, builds the order and carries it through to delivery.

    Taking real orders every day.

    GPT-4o tool calling, WhatsApp, Python

  6. Retail vision analytics

    Face and sentiment analysis at store exits to understand how customers leave feeling, and vehicle tracking that runs on small devices on site rather than in the cloud.

    OpenCV, Raspberry Pi, edge computing

What I do

AI that earns its place
Language models, document search, agents and computer vision, built into real products where they save time or make money, and tested like any other software.
Products from idea to launch
Web and mobile apps taken from a first conversation to a working release: scope, design, build, ship.
Backend and data
APIs, data pipelines, databases, background processing, and servers that stay up.
A second opinion
Feasibility, architecture, due diligence and build-or-buy decisions, before money is committed.

Tools I work with

AI
Large language models, RAG, agents and multi-agent systems, LangGraph, vector databases, computer vision, diffusion models, fine-tuning
Building
Python, FastAPI, Node.js, React, Flutter, WebSockets
Data
PostgreSQL, DuckDB, SQLite, MongoDB, Redis, BigQuery
Running it
Google Cloud, Docker, Linux servers, Cloudflare, CI/CD

Background

  1. 2021 - now

    Independent consultant

    More than seven AI products designed and delivered, for clients in India, the UK, Europe and the Middle East.

  2. 2019 - 2021

    Technology Head, EzCliniq Healthcare

    Healthcare mobile apps, language-processing systems for medicine queries, and research into reading handwritten prescriptions. This is where the patent came from.

  3. 2007 - 2018

    Director, then Senior Director of Engineering

    Crimson, The Advisory Board Company, then Optum. Built an engineering division from zero to 150 people and ran budgets of USD 5-7 million. Cut release cycles from 21 days to 48 hours, saved USD 1 million a year in infrastructure, and led the technical due diligence for the Optum acquisition.

  4. 2000 - 2007

    Engineering roles

    Tech lead on hospital systems at iSOFT in the UK, project leader at Covansys, web developer at Motorola in Singapore.

Patent holder for a digital prescription system for mobile devices. Degree in Mathematics from the University of Madras.

How I work

  1. A written scope and a fixed price

    You know what you are getting and what it costs before any work starts.

  2. Stages you can see working

    No long silences. Each stage runs, so you can try it and judge it yourself.

  3. Everything handed over

    The finished source code and the notes to run it are yours in full.

Have an idea? Let's build it together.

Tell my assistant about it below. It asks a few friendly questions and passes everything to me, and I reply personally within a day or two. It is an AI I built, so it is also a small sample of my work.

Prefer to write? Send me a message

Send me a message

Sent with basic browser details (browser, IP address, language, time zone, screen size, and the page that linked you here). Nothing is sold or shared.