I turn AI ideas into live products.
I'm Kumar Harsh, a freelance AI engineer. I architect autonomous AI agents, multi-agent swarms and custom RAG pipelines, plus the fast, intelligent apps around them, for founders and high-growth teams worldwide.
Selected work
Four of tenHoney
A private, witty, always-on assistant that lives in WhatsApp. It remembers, sets reminders, logs receipts, ghostwrites LinkedIn posts and understands voice notes.
- Category
- Autonomous WhatsApp agent
- Stack
- Cloudflare Workers, D1, Gemini 2.5 Flash
- Cost
- ₹0 a month
Memory
A persistent cognitive layer for the web. Attach thoughts, notes and tasks to any URL and watch them resurface when you revisit the page.
- Category
- AI agents and intelligence
- Stack
- Gemini 2.5, React 19, vector memory
- Live
- memory.kumarharsh.in
Point of view
I bridge the gap between cutting-edge LLM research and production software: building reliable autonomous agents, low-latency streaming interfaces, and resilient data pipelines that founders can scale
How I can help
Four services, each with a typical timelineAutonomous AI agents and swarms
Turn manual operations into self-directed AI workflows, with multi-step reasoning, external tool calling, browser actions and reliable execution.
Typically 2–3 weeksRapid 1–2 week AI MVP sprint
Zero to one: turn your idea into a production-ready, beautifully designed web app, complete with streaming LLM UX, auth, database and payments.
Typically 10–14 daysEnterprise RAG and context memory
Ground your AI in private documents, knowledge bases and user session context, modelled on the architecture behind Memory.
Typically 2–3 weeksAI reliability and chaos engineering
Eliminate rate-limit crashes, model downtime and flaky socket drops, stress-tested with the Hiccup chaos engine.
Typically 1–2 weeksFrom brief to production
My sprint, in four moves- i.
Scope
Tell me what you want to build. You get a scope and timeline estimate, with a direct reply within 12 hours.
- ii.
Architect
Agents, tools, retrieval and fallbacks are planned before the first line of code is written.
- iii.
Ship
A production-ready app with streaming UX, auth and a database goes live, with CI/CD ready on day 14.
- iv.
Harden
Guardrails, token budgets, observability and model fallbacks keep it running when a provider does not.
Endorsed by engineering leaders
Two LinkedIn recommendations“What sets Harsh apart is his practical knowledge of AI tools and agent orchestration, which he consistently applied to his front-end tasks.”
“He demonstrated strong proficiency in React, modern JavaScript, and component-driven architecture, consistently delivering high-quality, maintainable UI solutions.”
The toolkit
My core production stackWays to work together
A rapid zero-to-one prototype or a critical feature.
A full enterprise-grade build, with docs and hardening.
Ongoing fractional AI systems architecture and continuous shipping.
Writing
Notes on building AI systems that stay upLet's build something exceptional.
Have an AI agent, RAG pipeline or full-stack MVP to ship? Let's talk scope, architecture and timeline. I reply within 12 hours.