How we build
The domain is the requirement.
Our engineering philosophy fits in a sentence: software should understand the work, not just display it. Everything below is what that sentence costs to actually deliver.
01 · Depth before code
We learn the undocumented truths first.
Which ayanamsa convention a tradition expects. How Tally really rolls up a P&L — and why re-summing the leaves drifts from it. That a job number must continue a decade-old accounting series. That approval routing changes above a rupee threshold. None of this is in any specification; all of it is in our products. When we can’t learn a domain from outside, we operate inside it — one of our platforms runs a live business every day.
02 · Correctness, proven
Trust is engineered, not asserted.
Our astrology engine is validated against the reference implementations professionals already trust, re-verified by an automated 16-section regression suite. Our accounting answers match Tally to the rupee by using its own rolled-up balances. Our alerts re-derive their numbers through independent code paths before sending. Where certainty isn’t possible, the software says so — missing data is typed as unavailable, never fabricated as zero.
03 · Meet the world as it is
We integrate with what industries actually run.
Legacy desktop astrology exports with undocumented encodings. Tally on a single office PC, reached read-only with no inbound ports. WhatsApp as the interface, because that is where the user already lives. Offline-first field capture, because coverage fails. Zoom recordings that publish themselves into course chapters. The unglamorous last mile is where our products win.
04 · Language is architecture
Hindi and Hinglish are structural, not cosmetic.
Our products speak the language their users think in — a fully Hindi professional platform, Hindi as a curriculum axis, a deterministic Hinglish parser, a voice agent that converses in Hindi. This isn’t localisation applied at the end; it shapes parsers, prompts, PDFs, fonts and information architecture from day one.
05 · AI on merit
Intelligence, only where it earns its place.
Two of our products contain no AI at all — on purpose. Where AI runs, it is constrained: whitelisted tools, pre-computed numbers, server-side scope, independent verification, hourly grading against ground truth. The domain engine does the math; the model does the conversation.
For the technically curious
The stack, briefly.
Python/FastAPI and Laravel services on managed cloud infrastructure in India; React and React Native/Expo frontends; MySQL with Redis; Swiss Ephemeris for astronomical computation; object storage with CDN delivery; WhatsApp Cloud API, telephony, payment-mandate and video integrations as lean direct clients; realtime speech and language models applied narrowly.
The stack is deliberately boring. The interesting part is what we encode with it.