Turns public restaurant data into a structured intelligence report in about five minutes, compressing two analyst-days of research. Deployed on Hetzner; one pilot report walked through with a restaurant owner in New Delhi.

• Eight specialist AI agents (competitor, financial, review, menu, hypothesis verifier, validation, synthesis, owner narrative) over 8 public data sources. Scoring stays in code; the LLM does research and drafting.
• Three delivery modes: Public Scan, Validated Report (owner financials) and Owner Walkthrough (decision deck), each producing a ranked action plan, competitor benchmark and hypothesis checklist.
• Built end to end: Django 5 REST API, React 18 + Vite, PostgreSQL, Celery + Redis, WeasyPrint PDF reports.

Job matching and trust scores for informal workers in residential communities. Residents post jobs, get an AI-assisted price and pay workers directly via UPI. The platform never touches money.

• Django 5.1 backend on Railway, a Telegram bot for worker onboarding and job alerts, and one web portal for residents, workers and admins. Visual-first design for low-literacy users.
• Claude API for job-intent parsing and pricing recommendations from job category, location and historical rates.
• Pilot at Blueberry Homes, Bangalore, run with on-ground partner Alok; SOPs, job lifecycle and dispute handling documented.

Review workflow for Luxembourg AML/KYC teams: four-eyes approval, immutable audit trail and CSSF-format export. Password-gated demo at reasontrail.vercel.app.

• Three modules on a shared FastAPI + PostgreSQL backend: RiskRationale (risk classification with CSSF-mapped evidence fields), ReviewMemo (11-section periodic review memo + audit ZIP export) and AlertDesk (alert disposition with four-eyes approval).
• Designed the AI-assisted drafting layer: the model drafts, the analyst decides and signs, and every determination stays auditable by a regulator. The demo runs on a seeded synthetic scenario (Meridian Holdings).

Multi-agent personal advisor across nine life domains (finance, health, career, social, growth, library, travel, inner self, mindset). All LLM inference runs locally via Ollama; personal data stays on the device.

• Python + Streamlit app with multi-agent orchestration, DuckDB for structured personal data and one advisor interface across all domains.

Drone photography business infrastructure built end to end: portfolio site and print store, operations console, mission intelligence and a creative pipeline. Commercial launch pending business registration.

• Next.js 15 site and print store (Neon Postgres + Drizzle) with Stripe checkout, Gelato print-on-demand, DHL tracking and digital image licensing.
• Canonical data layer (VSIN registry) plus a Django 5.2 console on Railway: order ops, lifecycle emails, LLM dashboard and vision-model matching against the product registry. Console changes reach the site via ISR, no deploy needed.
• Vantage (vantage.vyomanaerials.com): LLM mission intelligence built in 3 days, with streaming briefs, airspace WMS checks, weather scoring and FLUX shot references.

Converts any PDF or EPUB into a narrated listening experience. Built for daily personal use, designed for anyone facing barriers to reading.

• 9-rule text preprocessing (punctuation repair, citation stripping, caption removal), AI chapter titling, adaptive buffered TTS, lock-screen captions and AI cover art. Cloudflare R2 as the only data layer.