Selected Work

We run our own business on what we sell

7 systems — products, our own internal tools, and client work. Each one explains, in plain language, the problem it solves and what it actually does.

We do not name clients, and we do not publish outcome percentages we cannot evidence. Every figure below is checkable against something real.

01

PropDash AI

Real estate CRMLive demo2026

Why it was built

Property enquiries arrive faster than an agency can qualify them, and the ones that go unanswered for an hour are usually gone. Agencies were tracking them in spreadsheets nobody trusted.

Evidence on record

Live demo you can actually click through — real AI answering, not canned responses

Role

Designed and built end to end, AI included

Stack

Next.jsTypeScriptVercelNVIDIA NIM

Delivered

  • Every enquiry scored automatically, so agents call the hottest leads first
  • A WhatsApp assistant that asks buyers the qualifying questions for you
  • Maps showing which localities and budgets your demand is coming from
  • Built so adding a second agency is a signup form, not a rebuild
View it live →
02

Lead Hunter

Finds your next customersInternal product2026

Why it was built

Buying a lead list gives you names, not customers. We needed a system that finds businesses that actually match what we sell, digs up a working phone number and email for each, and skips everyone we already know.

Evidence on record

Checks 12 different sources every night, automatically

Role

We built and run it for our own sales

Stack

PythonApifyNotion APIPM2

Delivered

  • Searches 12 sources — directories, maps, industry sites — while you sleep
  • Finds and verifies real contact details for each business
  • Scores every lead so you know who to call first
  • New leads filed into your CRM every morning, no typing
03

Content Engine

Writes and posts for usInternal product2026

Why it was built

AI that writes your marketing will happily publish something wrong — an invented number, a claim you never made. The hard part is not the writing, it is the checking. So we built the checking.

Evidence on record

5 separate AI checks before anything is published, running on our own LinkedIn today

Role

It writes and publishes our own LinkedIn content

Stack

PythonNVIDIA NIMOllamaLinkedIn API

Delivered

  • One AI writes, a different AI criticises — no AI grades its own homework
  • A fact-checker that blocks any claim it cannot verify, run twice before posting
  • Confidential material stripped out before any AI ever sees it
  • Posts on schedule without anyone touching it
04

The Watchdog

Catches silent failuresInternal product2026

Why it was built

The scary thing about automation is not that it breaks — it is that it can stop quietly and nobody notices. Ours once stopped for six days while every monitor said "healthy". So we built the watchdog that actually notices.

Evidence on record

Watches 9 of our own systems, checking every 5 minutes

Role

It guards every automation we run

Stack

PythonPM2Telegram API

Delivered

  • Notices within minutes when a job that should have run, didn't — and runs it
  • Catches jobs that are stuck, not just jobs that are dead
  • Never runs the same job twice by mistake
  • Knows customer-facing work must never fire at 3 AM
05

Automation Rescue

Fixed a broken automationClient engagement2026

Why it was built

A client's automation ran every day and quietly produced wrong results. Nobody could say which step was broken, so nobody could say what a fix should cost.

Evidence on record

6 separate problems found and written up, with two priced ways to fix them

Role

Found what was wrong, priced both ways to fix it

Stack

n8nWebhooksGoogle Sheets

Delivered

  • Walked through the whole automation, step by step
  • Every problem documented in plain language: what breaks, and what it costs you
  • Two honest options: patch what exists, or rebuild it with proper alerts
  • A written report the client could act on with us or without us
06

MediGuide

Healthcare, built for IndiaPrototype2026

Why it was built

Type a symptom into most health tools and you get American medicine brands that mean nothing at an Indian pharmacy counter.

Evidence on record

Built on Indian medicine brand names, not a US drug database

Role

Built the working prototype end to end

Stack

PythonRxNormNVIDIA NIMEdge TTS

Delivered

  • Describe the symptom, get the medicine
  • Answers use brands actually stocked at Indian pharmacies
  • A narrated demo video produced for evaluation
07

Trading Systems

Where we learned honestyInternal product2026

Why it was built

A trading bot that looks profitable on paper usually is not once real costs are counted. We built the measurement honest enough to tell us our own system was losing — and it did.

Evidence on record

595 real trades logged and measured, every cost counted

Role

Built the system, then built the proof it wasn't working

Stack

PythonSQLitescikit-learnWebSockets

Delivered

  • Trades measured against live market data, not a flattering backtest
  • An AI model that cannot beat a coin flip gets ignored, not trusted
  • Fees and costs counted properly — which flipped 'profitable' to 'not yet'
  • The discipline from this is in every AI system we build for clients

What we build on

We pick per project rather than forcing everything through one stack.

Next.jsReactTypeScriptTailwindPythonFastAPIPostgreSQLSQLiten8nWhatsApp Cloud APINotion APIVercelPM2NVIDIA NIMOllama

Have something like this to build?

Fifteen minutes is usually enough to tell you whether it is worth building, roughly what it takes, and whether we are the right people for it.

Book a free 15-minute call →