Privacy-Preserving AI Architecture: Federated Learning, Secure Aggregation, and Confidential Computing
Building intelligent applications around sensitive information creates a fundamental engineering...
Tutorials
Prompt Injection Through Documents: How to Harden an LLM Extraction Pipeline
A document is not a passive object once a language model reads it. Anything printed, hidden or...
How I ship fixes to open-source projects maintainers actually merge (with AI in the loop)
A failing test first, the smallest diff, and each project's AI policy respected: the workflow behind my merged open-source PRs.
Karpathy built HackerTube in 2 hours. I rebuilt it in the open and measured it
Inspired by Andrej Karpathy's HackerTube. Live demo: https://vishalmysore.github.io/OpenTube/ Hacker...
I don't read most of my AI's pull requests. I look at the ones I can see.
I don't read most of the pull requests my AI writes anymore. I read the title, the description,...
Running an issue-fixing agent on a $6 Fly Machine: a queue, three budgets, and an honest test report
Most "AI fixes your issues" demos are a session: you run a CLI, it edits files, you look at the diff....
When Should an LLM Router Stop Being Sticky? llm-d 1.0 Cache-Affinity Routing, Break-Even Math, and a Simulation
llm-d 1.0 routes by prefix-cache affinity until a token-load limit. Break-even math, a 10-endpoint simulation, and how to calibrate your own limit.
Building an Agent Evaluation CI/CD Pipeline: Architecture, Budgets, and Regression Gates
A complete engineering blueprint for running autonomous agent evaluation in CI/CD: tiered test runners, statistical regression gates, token budgets, and GitHub Actions orchestratio...
From 30 Minutes to 2: Building a Multi-Agent LangGraph + RAG Pipeline for Broker Back-Office Review
The problem: 30 minutes per application, 400 times a day On paper, broker back-office work...
My self-modifying system tried to wire a gate-passing module into its own protection enforcer. It was refused three times.
Part 4 of a series on a self-modifying multi-agent LLM system. Report and data:...
10 Open-Source Robotics Repositories for 3D Perception, VLA Policies, and Sensor Calibration
An inspectable technical breakdown of 10 open-source robotics and physical AI GitHub repositories solving training-free 3D occupancy, targetless calibration, 50Hz VLA latency, and ...
I used to dream about coding on the train. Now I'm managing agents from my phone. 🚉🍁🍂
This weekend I had the opportunity to work with v0, Claude Projects, Claude Code, agent skills,...
Discord Server Web Pages Bring Public Community Content to Search
Discord is expanding the web visibility of its Discoverable Servers through public Server Web Pages...
The Log That Told Everyone the Password: Redacting Secrets in n8n Execution Logs
The Log That Told Everyone the Password: Redacting Secrets in n8n Execution Logs I found...
I red-teamed my own support bot with 30 scenarios. It failed 7 — here's the harness.
Unit tests and eval sets don't catch the failures that end up in front of a regulator. Here's a 30-scenario red-team harness you can run against your live bot tonight.
I built a 3D game about my Nigerian childhood (and I'm not a game developer)
If you grew up in Nigeria, there is one sentence that can stop your heart in the middle of...
When I Was a Child, Mysterious Entities Used to Visit Me!
Hey everyone 👋, This post is a little different from everything else I have written on this blog....
45.8s, then 12.6s: same prompt on Cloudflare Workers AI
I've been building a small tool that turns one technical post into an X thread, a LinkedIn post, a...
AI Coding Is Scaling Faster Than Your CI Pipeline. Here’s What I’d Change
AI coding agents are increasing the volume and speed of software changes. This practical guide explains how to redesign CI feedback loops so agents can work faster without sacrific...
Your AI Feature Can Be Hijacked by the Text It Reads: Prompt Injection for Web Developers
You add an AI feature to your support tool. It reads each incoming ticket, writes a summary, and can...
The fix for an empty state isn't filling it
Last week I argued my cost tool should ship with an empty price book. Seven people told me how to make that less painful. Here is what I'm taking, and the one idea I'm saying no to...
Open Source Didn't Ban AI Code. 83% of Its AI Policies Allow It, With One Catch
A study of 281 AI policies in top GitHub projects: 83.3% allow AI code, 14.9% forbid it. What they reject is the diff nobody read. Review before the PR.
Pair agent UI work with visual regression before you trust green CI
Unit tests pass. Typecheck is green. Playwright happy-path clicks through. The agent PR still ships a...
I capped my coding agent's tool output to save tokens. It used 2.65 times as many.
Claude Code token usage, measured: a 12,000-character cap on tool output took one answer from 267k to 708k billed input tokens. Why, and what 252 runs showed.