We build software in Ruby and run it on our own machines.
Open, portable and self-hosted where it makes sense, with every outside service behind an adapter so no single provider owns us. Our engineering notes live at ruby.im.
HackRF, RSP1 and the limits of signal research: TEMPEST, cardiac sensing and a proposed authentication design, with an evidence ledger and companion essay on consciousness.
Beyond radio: programmable instruments, ranked: an editorial S+–F guide to electronics, acoustics, light, radar and experimental sensing.
Source discovery and collection priorities: an offline Ruby prototype with separate target/tool rankings, permission gates, evidence coverage and research handoffs.
Fortellis feature and problem catalog: a growing cross-department map of vendor claims, product scope and unresolved evidence, with search and open exports.
US dealer groups and rooftop websites: 16 group/business profiles and 15 sampled service rooftops or campuses, with scoped website/vendor evidence, proposed versus stated buying requirements and open comparison/export tools.
US OEM service messaging programs: 45 source-linked program, vendor, dealer-example and direct-app records, with historical dates, explicit unknowns, comparison tools and JSON/plaintext exports.
Dealership service messaging: source-backed product research, Ford and Tekion relationship checks, feature filters, side-by-side comparison and open data.
A controlled comparison of AI harnesses: a proposed system for testing harnesses and agent files with paired tasks, independent scoring, cost, safety and uncertainty.
Automotive marketplace research: searchable vendor-source product assessments for CDK and Tekion, with detailed profiles, coverage limits and open exports.
AI business models that can endure: YC and Palantir, forward-deployed engineering, restaurant and dealership opportunities, and a practical path from research to paid pilots.
Fortellis, feature by feature: every feature of CDK's dealer API platform, its 36 public APIs with listed prices, and all 479 marketplace apps from 421 vendors, classified by feature and AI use.
CRM and dealer systems for AI builders: every common feature of enterprise CRMs and dealership management systems, and how to build the APIs, MCP servers, data models and events other AI builders in automotive retail will consume.
A Rails stack on Rackspace Spot: deploying Ruby, memcached, Redis, Elasticsearch and Postgres on a Terraform-built Spot cloudspace, which services can ride preemptible capacity, and where data is at risk.
What is an AI agent harness?: 17 component responsibilities, concrete examples, a Ruby bug-fix walkthrough and a practical reading path.
Andrej Karpathy: a field guide for JayAI: neural networks, LLMs, agentic engineering, LLM Wiki and autoresearch, with critical analysis and an evidence-led roadmap for our network.
AI harnesses: what they are, how much of Y Combinator's 2026 batches they make up, and what may come after them.
Graph engineering: wiring agents, tools and knowledge into graphs, how many 2026 YC startups do it, and what may come after it.
A second brain on one laptop: the viral "Obsidian + Opus 5.5 + Jev" diagram checked against its sources, and the options for running each of its layers on our own machine.
REA and trading systems: the viral "most dangerous MCP for traders" post and its "hedge fund" paper checked, with what the six-step workflow can really do, why backtests lie, and where the legal line sits.
AI and day trading: what the evidence says about day traders and AI models trading real money, where real edges come from (unusual data feeds and prediction markets), stop-loss and bracket orders, and which broker API we'd use.
Money as a goal for AI: who is earning from AI in 2026, which business models hold up as intelligence gets cheap, what agents did when told to make money, and what we could sell.
Radio scanner: sweeping the spectrum with a software-defined radio and turning voice channels into text with local Whisper, lawfully and without root.
Radio station proposal: how the scanner grows into a station that charts activity, records ham nets, logs check-ins and traffic, and runs from a web page.
What this system is: JayAI, our self-hosted AI platform on one laptop, with its agents, sites, radio and camera apps, observability and the principles that keep it ours.
Hourly research with dot: preserve goals, evaluate recurring work, distinguish configured workflows from measured results, and draft version-specific feedback locally.
How our agents work together: ownership, isolated work, handoffs, independent review and controlled releases for Codex, Claude Code, Kimi and other agents.
Visitor and scan logging: counting visitors and spotting vulnerability scans on our own server, with a self-hosted dashboard instead of an outside analytics service.
Handoff README: how our AI server is set up, what runs on it, and how to add or change sites and pages.
Provided as is, for information only, with no warranty and no liability. Not legal or professional advice. Security and radio material is for authorized, lawful use only. Full disclaimer.
Visits are logged on our own server: IP address, browser and device details, and pages viewed. We use them to understand traffic and spot attacks, and never share or sell them.