Narasimha Badrinath

I build AI agents that survive the real world.

I'm a Staff Developer Advocate at DigitalOcean. I build AI apps and agents, contribute to the open-source tools developers use to ship them, and write about what I learn.

See my work Read my latest writing

Selected work

Voice assistant for macOS

Nix

Nix lives in my Mac's menu bar. I say "Hey Nix", ask something, and it answers out loud. The thinking is done by whichever coding agent I'm signed in to (Claude Code, Codex or Antigravity), so it can read files, run commands and use my tools, within permissions I set. Speech recognition runs on the Mac with Whisper, five times faster than a hosted service (265ms against 1.3s), and the audio never leaves the machine.

500+ commits · Swift app with a local Node service

How it works

Agent security

GuardianAI

Watches every tool call an agent makes and checks it against eight rules based on the OWASP LLM Top 10, covering prompt injection, leaked secrets, data exfiltration, SSRF and dangerous shell commands. Any FastAPI or Express app connects with two lines of code.

Open source · Python and TypeScript SDKs

Live demo Code Python SDK TypeScript SDK

Benchmark

Agent Memory Showdown

I benchmarked four LangChain memory backends over a 50-turn conversation. Vector memory kept 82% recall at about a tenth of the cost of keeping the full history.

5 runs per backend · Claude for generation, MiniLM embeddings

Code
Recall at turn 50 and cost per 1,000 conversations. Highlighted rows are the best value.
BackendRecallCost
Buffer90%$4,303
Qdrant (best value)82%$418
pgvector (best value)82%$416
Summary42%$2,256

Open source I maintain and contribute to

Latest writing

Sep 2026

Jev Is Fast Enough to Check Every Tool Call Your Agents Make

TypeSafe shipped Jev in September 2026. It cannot write text. What it does instead is answer a list of questions about a piece of data, with probabilities attached, in about a…

AI agents Security
Read
Jul 2026

Finding My Parents in a 24-Year-Old Scanned Voter Roll

It still baffles me how many people use AI today only as a fancy autocomplete, when the thing in front of them can do so much more.

Claude Code AI agents
Read
Jul 2026

Serverless vs. Dedicated vs. Self-Hosted LLM Inference: When Self-Hosting Actually Gets Cheaper

I ran the same model three deployment ways on DigitalOcean and measured the real costs, latency, and gotchas to find exactly when self-hosting stops being cheaper than serverless.

LLM inference Cost
Read
Jun 2026

When Your LLM Gets Retired: What You Actually Need to Do

A while back I got an email from OpenAI. A model I had been running in production, one I had carefully prompted, tested, and tuned, was being deprecated, with 60 days' notice. I…

LLMs
Read
All posts on the blog All articles Videos on Distilled AI

Live apps and side projects

AI projects

  • ArchitectAI LiveDescribe an app and watch agents plan, generate and check a cloud architecture for it.
  • Dead Startups Roast LivePick a failed startup, get it roasted by AI, then see how you'd rebuild it today.
  • Voice Course StudioAgents turn a raw recording into a finished explainer video, stopping three times for my review.

iOS apps

  • DailyBrickOffline habit tracker with streaks and widgets. No accounts, no subscription.
  • Silly DebatesPeople argue absurd topics for 24 hours, then AI reads every opinion and picks a winner.
  • FlipItCard-matching game with five themed decks and global leaderboards.

Speaking

Narasimha Badrinath on stage, presenting on running AI/ML workloads on DigitalOcean
Running AI/ML workloads on DigitalOcean

I give talks and run hands-on workshops on AI agents and taking LLM apps to production. I also review agent architectures for a few teams.

I'm always up for a meetup, conference or workshop. The easiest way to reach out is below.

Let's connect.

I'm always happy to talk about AI agents, developer tools, talks and workshops. Book a time, or pick whichever channel suits you.