An end-to-end, AI-native aerial-photography business — public site, print store, ops console, Vantage mission intelligence, and AI creative pipeline — designed, built, and deployed solo, with real payments, fulfilment, and customers.

• Public site + print store on Next.js 15 (Neon Postgres + Drizzle): Stripe checkout, Gelato print-on-demand fulfilment with DHL shipment tracking, returning customer checkout panel, and four-tier digital image licensing.
• Built a canonical data layer (VSIN registry) as the single source of truth, plus a Django 5.2 control-plane console on Railway — order ops with live Gelato sync, lifecycle emails, LLM intelligence dashboard, and VLM inference matching against the product registry. Console changes propagate via ISR — no deploy needed.
• Built Vantage (vantage.vyomanaerials.com) — 12-phase LLM mission intelligence in 3 days: streaming LLM brief generator (SSE), airspace WMS compliance, weather + drone suitability scoring, shot references via FLUX image generation + Pexels, and portfolio ROI attribution linking flights to orders.

LLM-powered compliance intelligence platform for Luxembourg AML/KYC teams — generates CSSF-inspection-ready review documentation with structured evidence citations, audit trails, and four-eyes workflow. Full MVP built in 2 days. Demo live at complylens.vercel.app.

• Three product pillars on a shared FastAPI + PostgreSQL backend (Railway): RiskRationale (risk classification with CSSF-mapped evidence citations), ReviewMemo (11-section periodic review memo + CSSF audit ZIP export), and AlertDesk (alert disposition with four-eyes approval workflow).
• Claude LLM drafts all compliance reasoning with structured evidence citations; humans review and sign — the design constraint is "auditable by a regulator," not just "LLM-assisted." Regulatory timing advantage: AMLR applies July 2027.
• Full password-gated live demo (Meridian Holdings synthetic scenario) with demo-reset endpoint; Next.js 15 + Tailwind 4 frontend on Vercel. Pre-seed exploration with co-founder Sachin Bhalerao; targeting Luxembourg ManCos, AIFMs, and compliance boutiques.

B2B SaaS platform that converts publicly available restaurant data into structured intelligence reports in under 5 minutes — compressing 2 analyst-days of research into a single automated pipeline. Deployed on Hetzner with a Vercel frontend.

• Built end-to-end: Django 5 REST API, React 18 + Vite frontend, PostgreSQL, Celery + Redis async pipeline, and a 6-agent LLM pipeline (scraping 8 public sources per restaurant).
• Three delivery modes: Public Scan (public data only), Validated Report (owner-uploaded financials), and Owner Walkthrough (5-page decision deck) — each producing ranked action plans, competitor benchmarks, and hypothesis checklists.
• Ran pilot reports for live restaurants; built a Validation Center to track which hypotheses owners confirm and a 30-day Experiments tracker for operational tests.

Trust infrastructure for the informal local economy inside residential societies. Workers list skills and build a verifiable trust score; residents post jobs, receive LLM-generated pricing recommendations, and pay directly via UPI. Platform never touches money.

• Built the full platform: Django 5.1 backend (Railway-deployed PostgreSQL), Telegram bot for worker onboarding and job notifications, and a unified web portal for residents, workers, and admins.
• Integrated Claude API for LLM-powered job intent parsing and dynamic pricing recommendations based on job category, location, and historical rates.
• Pilot-ready at Blueberry Homes, Bangalore — deployed to Railway with full environment setup; full SOPs, job lifecycle management, and dispute resolution documented in collaboration with on-ground partner Alok.

Privacy-first personal life intelligence system with six intelligence domains — finance, health, career, social, growth, and library — unified into one advisor chat interface. All LLM inference runs locally via Ollama; no personal data leaves the device. Live at soulspark.me.

• Built the full platform in Python + Streamlit: multi-agent orchestration, DuckDB for structured personal data, and a unified advisor pod serving six life domains from a single query interface.
• Privacy-first by design — zero cloud inference, zero subscriptions, zero data exfiltration. Uses Ollama with local models; all personal data stays on-device.

Personal audiobook system that converts any PDF or EPUB into a narrated listening experience — built as a product to use daily, and designed to work for anyone facing barriers to reading through vision, literacy, language, or attention.

• Narrate-Ready preprocessing pipeline (9 rules) produces clean, natural audio — punctuation repair, citation stripping, figure-caption removal, jargon normalization. AI chapter titling and a narrated brief (SummaryPlayer) surface context on every book.
• Cloudflare R2 as the data layer (no database), MediaSession API for lock-screen captions in three modes (sentence, line, chapter), DALL-E 3 cover art generation, and adaptive buffered TTS for instant playback. Built on Next.js 15 + OpenAI TTS.