Work

Things I’ve built and shipped.

AHA Smart Homes first, then earlier machine-learning work.

All case studies

Website · SEOAug 2026 – now

Website & search programme

Technical SEO, a content system and tracking that tells the truth.

For AHAMore new buyers find AHA on Google, without searching its name.

Web app · 3DSep – Oct 2026

Showroom TV app

A 3D tour, product demos and a pitch that plays itself, on one TV.

For AHAThe showroom TV now tells AHA’s story by itself, in about five and a half minutes.

AI agent · Data · BISep – Oct 2026

AI growth agent

Live CRM and ad data, checked twice, turned into a daily call list.

For AHALead and spend numbers AHA can trust, refreshed every 15 minutes.

Layers

GIS · MapsJul – Oct 2026

Deals map

Every AHA building on one map, with plans, billboards and the calendar.

For AHAEvery building with AHA homes on one map, and billboards planned around the leads.

growth_platform · Supabase · Claude

Data platform · AIApr – Sep 2026

Growth intelligence platform

CRM, calls, ads and maps in one place, with AI agents on top.

For AHAAHA’s first single view of leads, deals and sales calls.

Web scraping · GISOct 2026

Automation market map

Every automation firm in Mumbai and India, past the API’s 60-result cap.

For AHAA full list of who else sells automation, in Mumbai and across India.

Landing pageJun – Jul 2026

Luxury home-automation landing page

Fast, tracked, and wired straight into the CRM.

For AHAAd enquiries go straight into Zoho CRM, and the CRM keys stay off the page.

home_engine.py · rules + local LLM

AI prototypeFeb 2026

Smart-home command engine

Rules first, a local LLM second, guardrails always.

For AHAPlain-English home control that stays fast, offline and safe.

AI agent · DataFeb 2026

CRM daily insights agent

SQL computes the numbers; the LLM only narrates them.

For AHAA daily CEO briefing from the CRM, with every number computed in SQL.

Computer visionMar – Apr 2025

Live emotion detection

Seven emotions, live from a webcam.

Credit-card fraud · outliers flagged

Machine learningMay – Jun 2024

Credit score classification

100,000 messy rows to a 76.6% Random Forest.