I bridge engineering, content, and delivery.

Swaminarayan devotees needed spiritual guidance grounded in the Vachanamrut — but scripture is long and hard to search in the moment, generic AI chatbots invent citations, and the audience spans six languages on mobile. Wrong answers erode community trust.

• Scoped a lean PWA (no app store) with daily quotes, voice input, read-aloud, and WhatsApp sharing for how satsang communities actually share content.
• Chose Supabase for vector search and storage, Vercel for zero-ops deploys, and a split AI strategy: Gemini for embeddings and speech, GPT-4o for structured spiritual answers.
• Defined trust guardrails — citation stripping when scripture doesn't cover a topic, automatic language retry, layered rate limits, and a humble "guide, not guru" tone.
• Pre-processed PDF scripture into a searchable vector knowledge base instead of querying live documents.

Solo product owner: I scoped the system end-to-end — ingestion pipeline, retrieval logic, API boundaries, multilingual prompts, and launch. AI-assisted development tools (Cursor) accelerated implementation so I could focus on architecture decisions, content guardrails, cost controls, and deployment — not syntax.

• Live at askmaharaj.com — 6 languages, voice in/out, installable PWA
• Daily curated quotes with IST rotation and community sharing
• Layered cost controls so the product runs on free and low tiers
• Operational review loop via structured Q&A logging

Families and mandirs relied on scattered PDFs, WhatsApp forwards, and multiple tabs for lyrics and meaning. Recordings lacked synced text for learning. Elders knew Gujarati and Hindi; younger members needed English context — with no single, respectful place to read, listen, and learn kīrtan online.

• Built a multilingual kīrtan library with verse-level karaoke, search and filters, and side-by-side translations across English, Gujarati, and Hindi.
• Designed a rich content model — verses, per-locale translations, karaoke timestamps, and human vs. AI tagging — managed through a super-admin CMS.
• Positioned AI as assistive only: line-level Ask and full-kīrtan Arth, sign-in gated, never presented as scripture.
• Shipped PWA with offline audio saves for home and mandir use; Panchang-based daily picks scoped for a later release.

Directed product scoping and AI-assisted delivery of the full platform — content schema, user journeys, admin workflows, and SEO. Used Cursor to accelerate UI and API work while I owned data modeling, karaoke sync requirements, auth and offline behavior, and Vercel deployment.

• Live catalog with karaoke player, favourites, and offline playback
• Super-admin CMS for publishing kīrtans, verses, and audio timing
• Multilingual UI with accessibility-first, sacred-content-first design
• Part of a product family alongside Ask Maharaj for satsang communities