Hey, I'm Ravi 👋

I build things. Lately, most of them think a little.

Bengaluru, India


I've been shipping software for about a decade — long enough to have a graveyard of early iOS experiments, a couple of Flutter apps people actually starred, and a growing pile of AI systems I'm genuinely excited about. I'm a generalist at heart: I go wide across mobile, backend, infra, and ML, and I learn by building real things rather than reading about them.

These days I'm most drawn to AI that has to earn its keep — agents that do real work, RAG that cites its sources, tools that make other engineers faster. I spend most of my time on the unglamorous parts: Does it actually work? Can you tell why it did that? Does it run without me babysitting it? That's usually the difference between a demo and something people rely on.

I also have a soft spot for tools that run on your own machine — no API keys, no phoning home.

Ask this portfolio runs in your browser · 0 API calls

Retrieval over this page plus a small language model, entirely on your device — nothing you type leaves your browser.

until then, answers quote the sources directly
agents welcome too: claude mcp add --transport http ravixdaisy https://ravixdaisy.com/mcp

🔨 What I'm building now

  • portico — self-hostable browser automation for regulated portals (healthcare payer / EHR), where PHI can't leave your VPC. The interesting bit: an LLM authors and repairs the automation, but never runs on the hot path — so a healthy run moves at browser speed, not model speed.
  • org-context — a context engine for coding agents. It ingests a team's code, PRs, tickets, and incidents and serves permission-aware, source-backed context over REST, MCP, and CLI. Basically: give your agents the same context a senior engineer has.
  • study_graph — a study buddy that runs entirely offline. Drop in your PDFs and notes, ask questions, get cited answers and flashcards — local models, local database, zero API keys.
  • lumen — a local-first desktop workbench for poking at, debugging, and operating Kubernetes clusters without living in kubectl.

There's also a patient-engagement recommendation platform I'm building privately — rules-first, ML-assisted WhatsApp outreach for a health team, where the model suggests but never overrides the safety rules, and every recommendation is fully auditable.

📱 Where I started

Before any of the AI stuff, I was a mobile dev. A couple of those apps are still my most-starred repos — which is either humbling or motivating depending on the day.


🧭 A few things I've come to believe

  • Ship it, then make it good. Something in production teaches you more than a perfect plan.
  • If the AI can't explain itself, it doesn't get to decide. Confidence without a reason is a liability.
  • Small and self-contained beats big and needy. I'd rather hand you something that runs on your laptop today.

🛠️ Usual suspects

TypeScript · Python · React · FastAPI · Node · PostgreSQL · Playwright · LangGraph · Ollama & local LLMs · pgvector · Docker · Kubernetes — and whatever the problem actually needs.


Always up for good problems — AI products, dev tooling, open source, or anything worth building well.

Let's talk. 📍 Bengaluru, India