Microsoft's open-source method for training agent skills like a neural network, without touching the model: small edits to a skill file, kept only if they score better on held-out tasks.

An open-source framework that analyses a stock the way a trading firm would: analyst agents, a bull-versus-bear debate, a trader, a risk team and a portfolio manager.

An open-source Claude Code framework that runs a whole job search: profile, portal search, fit scoring, tailored CV and cover letter, reviewer agent, interview prep and tracking.

A cloud drive that shows up in Finder but takes no disk space: files stream on demand, every save is a version, and agents can work on instant copies of a whole drive.

NVIDIA's open-weight 100M-parameter model that tells who spoke when, for up to 8 speakers, live or recorded. Ranked first on VoiceArena's diarization benchmark at launch.

How I configure Claude Code for NicAI: layered CLAUDE.md files, hooks, auto-memory, permissions, skills, connectors and the output conventions.

What NicAI has cost me in time and services, what failed on the way, and the lessons I would give anyone building their own AI layer.

Why SQLite files, Markdown knowledge bases and JSON are the source of truth for NicAI, and the 4 nightly jobs that keep them current.

LlamaIndex's open-source local document parser, rewritten in Rust: PDFs and Office files to text, JSON or Markdown in milliseconds, with no model and no cloud.

My personal AI layer: what it is, why I am building it, how it started in 2022, and where it stands in September 2026.

NicAI Lab, the site where NicAI and I put the pages, demos and files we build, shared through unlisted links that only work if I give them out.

How NicAI turns "write a note about X" into a live, signed page on this site in about 15 seconds: the write-note skill, the publish hook and the 5 publish phases.

Anthropic's guidance for getting the most out of each Opus release in Claude and Claude Code, newest first: Opus 5.5 (September 2026) and Opus 4.7 (April 2026).

The 30 small Python wrappers my NicAI skills call for research, web data, media, publishing, email, DNS and home automation, and the pattern they share.

Every skill in my NicAI setup, grouped by purpose, 1 line each, plus what a skill is and the rules I build them by.

A conversational video avatar of me, built on Kaltura's Immersive Agents platform: what it is, how it is configured, and what I learnt.

How NicAI tries to write like me: stacked writing rules, a persona spec built from my own sent messages, a blind test, and why drafts still need my edit.

How my ~/ai folder is split into domains, the file conventions every agent follows, and how a plain TASKS.md file drives the agents.

A Rust terminal multiplexer built for coding agents. Sessions survive a closed lid, and every pane says whether the agent is working, blocked or idle.

Cloudflare's open-source skill that turns a coding agent into a security auditor: 6 phases, independent verification, and machine-readable findings.

Dwarkesh Patel's 80-minute interview with OpenAI's Noam Brown: how 10,000-agent swarms work, what maths progress says about RSI, and the alignment problem it raises.

A 10-day family trip to Japan in 2027, landing in Tokyo: the route, when to go, the best sushi, where to see master craftspeople, electronics, and the rule changes since 2026.

A repairable, open-source inkjet from Paris that refills standard HP cartridges, prints on roll paper and hides nothing behind DRM. Still before launch, with no price yet.

Pi is a minimal, open-source, bring-your-own-key coding agent harness, and the harness alone can cut cost about 2x for the same model.

Training a small model to write in my voice, wired behind a frontier model that does the thinking. When it beats prompting, and what it costs.