Blog
Measured studies, open-source implementations, and production lessons across edge AI, agent reliability, privacy, cost, and security.
Measured studies, open-source implementations, and production lessons across edge AI, agent reliability, privacy, cost, and security.
Training is moving back onto the machines people own, and those machines are not a fleet you control. Apple GPU support sat on my plan for two years because nothing looked broken; the runs finished, they were just slower. Turning Metal on was worth 1.7x to 4.2x per epoch on the one M1 I own, and much less end to end. The measuring is what found the real bugs, three of which never raised anything: is_bf16_supported() answering True on a card that only emulates it, a gradient scaler silently dropping optimiser steps and halving a mAP, and 16-bit being slower than 32-bit on Apple Silicon even though everything works.
Models almost never ask you what you meant: across ambiguous questions, answer rates run above 95% and clarification barely registers. Repairing that in conversation makes it worse: underspecified multi-turn runs measure a 39% average performance drop, driven by a 112% jump in unreliability, and the effect shows up from two turns onward. The leverage is no longer in writing better prompts. It is in forcing the questions out before the first answer attempt, and in asking the questions afterward that prove you understood what you shipped.
A Flatpak build that passed every structural check and still died one line into launch. Why green CI proved nothing about the GUI, and how to install the real artifact and start it under Xvfb instead.
WebMCP lets a web page hand callable tools to browser AI agents instead of making them scrape the DOM. I wired three read-only tools into this site. The interesting part was not the API. It was deciding what an agent, or a prompt injection riding one, is allowed to do.
When a machine writes most of your diff, the constraint stops being how fast you type and becomes how well you review. Here is the SDLC I actually run (the provenance audit, the sandbox, the CI backstops, and the three questions I ask every AI-authored change) plus the parts of review that don't compress and never will.
On small local models, the standard ReAct loop has a failure mode nobody warns you about: the model calls one tool, declares victory, and stops. What we measured across 12 GGUF models in EdgeVox, why we added a plan-once dispatcher, and how to decide which loop your task actually needs.
A first-hand build narrative of EdgeVox, a fully offline voice agent that chains speech-to-text, a local LLM, and text-to-speech on one device. The architecture in plain language, ROS2 integration, the latency budget, and the failure modes nobody warns you about.
Building a represent-me chatbot is a weekend project. Treating it like a production security surface is the part nobody writes about. Here is the architecture, the prompt leak I found by attacking my own bot, and the reusable suite that keeps it honest.
Vietnam already has the chips, three meaningful Vietnamese model attempts in flight, and the most binding AI law in Southeast Asia. The conversation about sovereign AI keeps demanding a 70B foundation model. The actual gap is one layer down: open evaluation, license-clean data, compliance-aware specialized models, and on-device runtimes that operationalize Law 134/2025 from March 2026.
A 2025 RCT found students who studied with ChatGPT scored 11 points lower on a retention test six weeks later, even though their work during the sessions was higher quality. Better results, worse thinking. Here is what the cognitive science says, and how it changed my own AI habits.
NVIDIA just open-sourced OpenShell, a policy-enforced sandbox for autonomous AI agents. Four security layers, a privacy router that decides which LLM sees which data, and hot-reloadable YAML policies. How it works and what it solves that nothing else does.
Vibe coding is shipping apps faster than ever, but also shipping vulnerabilities at alarming rates. This guide covers real disasters, practical security measures from secret scanning to endpoint hardening, and SOTA AI-powered review methods to keep your vibe-coded apps from becoming the next breach headline.