LoveHurts
Health platform
Built in Next.js and React, designed for HIPAA requirements with encryption at rest.
- Next.js
- React
- HIPAA
Sunny Luthra
AI product engineer and dog behaviourist. Full-stack (Next.js, React Native, Node.js) · LLMs, RAG and agents.
- Pune, India
- sunny@mrova.rocks
- GitHub
I have been writing code since 2004, when I built my first website. In 2011 I started mRova, where I build web platforms, mobile apps and AI systems for clients.
I also build and run my own products: Everything.Dog, an AI app for dog parents on iOS and Android, and Bos.Dog, software for dog-boarding businesses. I built a D2C brand that has served 5,000+ orders.
My recent work is on AI agents: open-source MCP tools in Rust and TypeScript, and LLM evaluation.
I am also a dog behaviourist. I have worked with Indian dog parents on behaviour problems since 2015, and I wrote Raising Well-Behaved, Balanced Dogs. I live on a hill near Pune with 13 dogs, 8 puppies, 2 cats, a calf and a couple of snakes.
- 2004
- First website
- 2011
- Started mRova
- 2015
- Dog behaviourist
- 2022
- Published a book
Experience
Jul 2011 – present
Founder & Lead Product Engineer
mRova · Product engineering studio
Pune, India
mRova builds web platforms, mobile apps, AI systems and custom software. The same team scopes, builds, deploys and supports each product. I lead client work from discovery and product strategy through design, build, deployment and support. Several clients have worked with me for years.
WazirX Research
Multi-language content platform
For one of India's largest crypto exchanges, with Algolia search and a multilingual CMS.
- Algolia
- Multilingual CMS
Shardeum
Developer-education site
For a Layer-1 blockchain, on a custom WordPress theme and plugins.
- WordPress
- Custom theme
Also
Healthcare and pet-care sites with Calendly booking, WhatsApp API and SEO: 32Smiles Dental, an ENT specialist, a senior pulmonologist, Dogmantra, PetAshram and Kukkur Vachan. CRM, analytics and marketing automation set up so each site works as a lead-generation system.
Jul 2016 – present
Founder & Product Engineer
OhMyDog Rocks · AI pet-care company
Pune, India
I founded and run a pet-care company as both the operator and the lead engineer. It combines a dog-behaviour practice with our own AI products. I handle product strategy, pricing, SEO and growth marketing.
Everything.Dog
AI companion app for dog parents · App Store and Google Play, v2.3
React Native with Expo; a Fastify and TypeScript API on PostgreSQL (Prisma, pgvector), Redis and Socket.io; RevenueCat subscriptions; Google and Apple sign-in.
- EVA chat
- Classifies each message into 14 intents and routes it to gpt-4o or gpt-4o-mini. Answers from a knowledge base (RAG), keeps long-term memory of each dog with pgvector and rolling summaries, takes voice input (Whisper) and calls about 14 tools. Sends push messages from 8 triggers on Bull queues.
- Image analysis
- 8 analyzers on GPT-4o vision with structured output: body condition, breed, mood, poop, skin, X-ray, blood report and food labels. Results show as cards in chat and as PDF reports.
- Walks
- GPS tracking with background location, sniff-stop detection, streaks, live walk sharing and a map-tile territory game with leaderboards.
- Admin
- A Next.js dashboard for reviewing EVA conversations, memory and emergencies, moderating the community and tracking AI usage.
- React Native
- Expo
- Fastify
- PostgreSQL
- pgvector
- OpenAI
Bos.Dog
Multi-tenant SaaS for dog-boarding businesses
Inquiry, booking, check-in and out, daily care logs, a parent portal, report cards and invoicing. English and Hindi, settings for six countries, about 2,000 automated tests, and CI deploys with health checks and automatic rollback.
- Next.js 15
- PostgreSQL
- Drizzle
- BullMQ
- Razorpay
AI community assistants
Six persona bots for a Telegram dog-parent community
They answer from a curated knowledge base, and a person approves every reply. Evaluation sets built from about 1,260 past threads; six embedding models benchmarked on English and Hinglish questions (the chosen model reached 74% recall@5, against 62% for keyword search); an LLM-judge replay harness picked a lower-cost drafting model.
- TypeScript
- grammY
- sqlite-vec
OhMyDog.Rocks
AI-powered dog-care platform
Built on Next.js and OpenAI, with a Telegram channel for client conversations and a searchable archive of transcribed Hinglish coaching calls.
- Next.js
- OpenAI
- Telegram API
OhMyDog Academy
Online dog-behaviour education
Launched on WordPress, Tutor LMS and WooCommerce.
- WordPress
- Tutor LMS
- WooCommerce
Dog behaviour
I founded OhMyDog.Rocks as a behaviour practice. I am not a vet: I read what a dog is doing and why, and hand clinical decisions to a vet.
Book
Raising Well-Behaved, Balanced Dogs
The Dog Parent's Guide Series
A book about dog psychology, not tricks or commands. It covers how dogs think, the daily rituals of dogs and how to perform them, common behaviour problems and their fixes, five skills to teach your dog, and the support system around a dog parent. Written for dog parents, and for trainers and behaviourists.
- Publisher
- Notion Press
- Published
- 6 April 2022
- Pages
- 44
- Amazon.in
- 4.9 / 5, 22 ratings
How it started
- 2015
Webby, my first dog, arrived. Every training school on offer taught obedience rather than understanding.
- 2016
Started observation sessions with other dog parents.
- 2019
Moved to a quiet hill near Pune to watch free-roaming dogs.
- 2020
Webby died of tick fever. Since then the work has put prevention, food and gut health first.
Consultations
In-home behaviour consultations across India. Psychology first, structure second.
Certification examiner
I set the standard for the OhMyDog Certified Dog Behaviorist programme and examine every candidate. The first batch filled at 30+ candidates.
Methods
Created the Rituals of Dogs (six daily rituals), the OhMyDog Parenting Model and the Calm Walk Method course on leash psychology.
Writing and products
Every guide on ohmydog.rocks carries my name. I also formulated ProBelly, a gut-health supplement for dogs.
Open source
Tools I build for working with coding agents. Both are MIT and on GitHub.
snyvi
Desktop app for coding agents
Up to four coding agents (Claude Code, Codex) work side by side on each project, and everything they write is kept as versioned, searchable pages. A 19-tool MCP server and end-to-end encrypted sharing through a Cloudflare relay. 49 releases in its first month for Linux, macOS and Windows; 424 tests; 11 ms cold start, with performance budgets enforced in CI.
- Rust
- SQLite
- Tauri
- MCP
snypd
A CMS that coding agents run over MCP
Markdown in git, static HTML out with 0 KB of JavaScript. Published to npm for five platforms, with 4 themes, 4 plugins and about 700 tests. Builds 1,000 pages in 2.7 s, rebuilds one edit in 13 ms, and deploys to Cloudflare in one command.
- TypeScript
- Bun
- MCP
AI and developer-tool projects
HarnessArch
A specification for AI agent harnesses
30 essays and 63 interactive diagrams.
- Next.js
- MDX
- React Flow
mrova.rocks
A company site built to be read by AI agents
A markdown copy of every page cuts what an agent has to parse by 92%, measured on 9 pages. Also serves WebMCP tools and llms.txt.
- Next.js
- WebMCP
- llms.txt
OhMyVet
Prototype: reading veterinary lab reports
Eval-first: every component had to beat a recorded baseline. 99.65% analyte recall on 38 generated test reports, at about $0.013 per document.
- Google ADK
- FastAPI
- pgvector
Skills
- AI and LLMs
- OpenAI and Claude APIs
- Vercel AI SDK
- RAG
- Agent memory (pgvector, sqlite-vec)
- Tool calling and model routing
- MCP servers
- LLM evaluation (golden sets, LLM-as-judge, retrieval benchmarks)
- Vision models with structured output
- Whisper
- Frontend and mobile
- React
- Next.js
- TypeScript
- Tailwind CSS
- React Native with Expo
- Background location, maps, push, in-app subscriptions
- React Flow
- MDX
- Backend and data
- Node.js
- Fastify
- Bun
- Rust
- REST and GraphQL
- PostgreSQL
- MySQL
- SQLite
- Prisma
- Drizzle
- Redis and BullMQ
- Socket.io
- Supabase
- Firebase
- Integrations
- WhatsApp Business API
- Telegram API
- Algolia
- Calendly
- Razorpay
- RevenueCat
- Shopify
- CRM and analytics
- CMS and commerce
- Payload CMS
- Headless and custom WordPress
- WooCommerce
- Tutor LMS
- Infrastructure and security
- Cloudflare Workers and Durable Objects
- Docker
- CI/CD with health-checked rollback
- Signed releases
- Tauri desktop apps
- End-to-end encryption (x25519, ed25519)
- HIPAA-oriented design
- Product
- Scoping
- Roadmaps
- Pricing
- SEO
- Conversion optimisation
- Growth marketing
Writing
- Harness EngineeringAgent failures are usually harness failures, not model failures. The eleven primitives that decide what an agent can see, do, remember and recover from — and how to find which one ran out of road.
- Making a Next.js Site Agent-ReadableServing a markdown twin of every page cuts what an agent has to parse by 92%. Serving an llms.txt may do nothing at all. Here is the code for both, and the evidence for each.
- WebMCP: Give the Agent a Menu, Not a MapWebMCP lets a web page register typed tools that an in-browser AI agent can call instead of scraping the DOM. The spec, both APIs, a Next.js implementation, the security rules, and whether to ship it now.
- Context Engineering vs RAGRAG gets material in. Context engineering decides what earns the room. The distinction sounds academic until quality starts falling as your context window fills.
Contact
Email is the fastest way to reach me.