How Each Product
Was Actually Built.

Developers building astrology applications had no reliable, comprehensive API to build on. Existing solutions offered 10–20 endpoints at best, relied on external ephemeris services, and had documented accuracy issues with edge-case chart calculations. There was no professional-grade, developer-first Vedic astrology API that could serve as a real foundation — not just a toy project.

The core decision was to compile Swiss Ephemeris directly into the runtime so all planetary calculations happen locally — eliminating third-party dependency for the most critical part of the system. From there, the architecture was built endpoint-first: every Vedic chart component (Lagna, Navamsa, Dasha, Shadbala, divisional charts) modelled as a separate endpoint with typed responses, not a monolithic query.

Claude AI was integrated for narrative generation — turning raw chart data into human-readable interpretations — and a streaming chat widget was added so developers could embed AI-powered astrology assistants without building that layer themselves.

Coaches and consultants were stitching together 4–6 separate tools that didn't talk to each other: one for booking, another for payments, another for courses, another for email. The monthly SaaS bill was significant — and on top of that, platforms like Teachable and Podia take 5–15% of every sale as a platform fee. The market needed one platform that consolidated everything with no tax on earnings.

The architecture was designed module-first: booking, payments, products, email, and CRM were each built as isolated feature modules sharing a single Supabase database. This let each module ship independently without coupling everything together.

The payment layer was built directly on Razorpay and Stripe — not through any intermediary platform — which is what makes 0% platform fees possible. The AI Marketing Suite was added as a separate module, using Claude to generate full marketing strategy outputs in under 90 seconds from a short business brief.

Everyday users had no accessible way to verify whether a file, link, or image was safe. Enterprise security tools required technical knowledge and paid subscriptions. Free tools were either inaccurate or sent your files to a server — which is its own privacy risk. Deepfake detection was completely absent from consumer tools. There was a clear gap for a privacy-first, accessible scanner that anyone could use without uploading their data anywhere.

The privacy-first constraint was non-negotiable from day one: files must never leave the user's device. This ruled out most server-side scanning approaches and pushed the architecture toward WebAssembly — running scanning logic directly in the browser with no round-trip to a server.

URL and phishing detection uses a combination of heuristic pattern matching and Claude AI analysis to flag suspicious content and explain why in plain language. Deepfake detection leverages browser-native APIs combined with AI analysis of image metadata patterns. The UX goal was: anyone should be able to use this without reading a manual.

People facing critical decisions — job changes, financial moves, major life timing — had no structured way to get clarity. Generic AI chatbots gave vague, hedge-everything answers. Career consultants were expensive and often generalized. What was missing was a focused AI tool that gave actionable next steps with timing guidance — not fortune-telling, not generic advice, but structured intelligence applied to a specific decision.

The product was built around a single insight: people don't need more information, they need structured clarity on what to do next. The UX is deliberately narrow — three decision areas (career, money, life), a short guided input flow, and a structured report output.

The AI layer uses a multi-stage prompt chain: raw inputs are transformed into a decision context document, which is then analysed through a timing and risk framework before generating the final report. Claude AI's long context and reasoning depth were essential for making outputs feel substantive rather than generic.