Client sites you can open and check, demonstration builds we made uncommissioned, and the internal systems behind them — a double-entry ledger, a CNC production pipeline, an outbound engine. Each case study covers the architecture, the stack and the decisions, and says plainly which of the three it is.
Saraswati Dental needed the enquiry captured before the visitor left the page — and routed somewhere a busy practice actually reads. The build pairs a validated intake form with a WhatsApp hand-off that arrives already knowing what the patient asked for.
Read the case study
Someone choosing a neuro-rehab clinician checks qualifications before price. SPARC needed those credentials stated where both a patient and a search engine could verify them — and an intake that reaches the clinic on the channel it already answers.
Read the case study
Plot buyers search corridors and landmarks, not developer names. R³S needed a site that answered locality queries directly, published its inventory in a form search engines can read, and moved an interested buyer to WhatsApp in one tap.
Read the case study
DemonstrationA demonstration build for a dental and diagnostic centre in Tundla, put together to show the practice what its site could be. The interesting constraint was delivery: the whole thing is a single HTML document with the clinic photography embedded in it, so it runs from any host — or from a USB stick — with nothing to install.
Read the case study
DemonstrationBuilt after the single-file demonstrations, and deliberately not like them. Every clinic-specific fact — name, address, treatments, hours, contact links — sits in one typed module, separate from the fourteen components that render it. Re-pointing the system at a different practice is a data edit, not a rebuild.
Read the case study4 source artifacts · src/ledger/engine.ts — 438 lines
Indian cash businesses already record their day in messages — “आज ₹25,000 cement kharida”. The hard part is not reading that sentence. It is making sure a machine that half-understood it can never write a wrong number into the books.
Read the case study6 source artifacts · orchestrator.py — 1,261 lines
A WPC door design has to exist in every size a customer might order, in four different file formats, each named exactly the way the CNC operator expects. Done by hand it is an afternoon of exacting, error-prone work per design — and a single mistyped size reaches the router as scrapped material.
Read the case study5 source artifacts · scripts/system_orchestrator.py — 204 lines
Most outreach tooling is good at sending and bad at listening, which is how a prospect who answered on Tuesday gets chased again on Thursday. This engine treats the inbox as part of the pipeline: replies come back in, and the sequence for that lead stops.
Read the case study5 source artifacts · src/server.js — 458 lines
Text agents get to think. A voice agent has roughly a second before the silence sounds wrong and the person on the other end says “hello?”. Everything in this build — the model choice, the speech cache, the turn handling — is spending that budget.
Read the case study3 source artifacts · parser.py — 1,222 lines
The statement arrives as a photograph of a page. Somewhere in it are three hundred transactions that a business needs in a spreadsheet — and typing them out is how a Saturday disappears. This pipeline reads the scan and hands back the workbook.
Read the case studyNo outcome figures appear on these pages. SheetFlow publishes a result only once it has been measured against a stated method and confirmed with the client in writing. Where that has not happened, the case study says Not publicly disclosed rather than estimating.
The audit maps where your enquiries arrive, where they stall, and which steps are worth automating first. No commitment, and you leave with the map either way.
