The portfolio · 23 products
The Autonomous
Systems Portfolio.
Each product owns a layer of how autonomous systems work — observing the world, predicting it, deciding, acting on it, and governing themselves. Together they describe an ecosystem, not a list.
Read it as a timeline, not a bundle
A portfolio is the wrong word.
There’s a paradox worth sitting with before you scroll. Everything below came out of a classical training — the slow route: years in the room, a mentor, a ladder, an industry that paid you for accumulated experience. Now take that route away and ask the honest question. Where do those opportunities lie for someone still in school? What is their job today, and what is it tomorrow? The path that produced people like me is quietly disappearing — the kind of past experience this was all built on barely exists anymore, and soon may not exist at all.
That’s why so many skilled people — engineers, marketers, whole professions — feel a real resentment toward tools like these. The resentment is rational: we all know that if some of these outcomes land, they mean a genuine shift in job types, in job roles, and a huge amount of disruption. It’s the same disruption that makes large organisations struggle, because AI transformation isn’t another software upgrade — it’s a whole new ball game. Those who realise that will flourish; those who don’t will struggle.
So this isn’t a catalogue of products, and it isn’t a passing thought on a page. It’s one continuous, documented response to that shift — every tool forced into being by a real constraint, in real order, for real reasons. Hot air sells; the walk is where most stumble. This is the walk: hard evidence, not a pitch — real code, shipped, and most of it deliberately reused from one build into the next. Tap any node for the thinking, the design rationale and the shared code behind it; read the full essay for the whole argument.
Read the full story the full essay — five years, in orderThe flagship · Release Ready
Silo — The layer your AI can't lie to.
One of the twenty-three carries the most weight. Silo is the first security architecture built to watch AI agents themselves — five independent layers from silicon to cloud, where compromise has to defeat all of them at once. It is the second moonshot in the portfolio, and the near-term one.
- The first security layer built to watch AI agents
- Five layers — silicon → hypervisor → OS → agent → cloud
- >90% detection accuracy · <1% false positives
- 1.8M lines of code · 10 papers (240+ pages) · 6 platforms
Decision Intelligence
Observe and predict — truth and foresight for high-stakes decisions.
SignalFabric
Decision intelligence, fused from every signal.
Institutional quant, made retail-accessible. It fuses 506 signal primitives across 16 quant categories and 40+ integrations — markets, macro, sentiment, prediction markets and physics-informed models — into explainable, calibrated forecasts, with a trust layer that scores manipulation. And it acts — describe a strategy in plain English and it can backtest and trade it live, guarded by an AI risk critic and a kill switch.
posit.
Post-trade truth for Solana traders.
Post-trade analytics for Solana traders — your trades distilled into clear scorecards and real on-chain signal, in a dark, fast interface built for people who want the truth, not the hype.
memedataengine
The full pump.fun firehose — decoded, retained, scored.
An instrument for the strange world of the degen — the entire pump.fun firehose decoded, kept forever, and scored for fraud, sentiment and smart-money signal. A read-only intelligence terminal that watches the chaos so you don't have to trust the hype.
Strategy as Software
Does the business behave like the plan you approved?
Strategic assurance for incoming CEOs, boards and operating partners. It reads the strategy you approved, reads what the business actually did — budgets, board packs, hiring plans, contracts — and shows you, with page-level citations, where the two have quietly stopped agreeing.
FAULTLINE
Contradiction intelligence for public companies.
A research instrument that tests whether what a company reports, what management says, what outside evidence shows and what the market prices are still mutually coherent — and surfaces the places where they are not. Short-side equity research is the first experiment, because a negative thesis is cleanly falsifiable.
Baha Manila
One screen, one watershed, the moment the alert lands.
A free flood monitor for the Marikina watershed — river gauges, dam bulletins and spillway status joined into a single pane of glass for the ten minutes after the national alert arrives. Official and derived kept honestly apart, and stale data admitted rather than hidden.
Governance & Security
Govern autonomous systems — monitor, test and verify what is real.
Silo
The layer your AI can't lie to.
The first security architecture built to watch AI agents — a five-layer defence from silicon to cloud that detects compromised, weaponised or rogue agents, with cross-layer detection that is physically, not just computationally, hard to fake.
Bertha Siege Engine
AI-driven penetration testing for the modern cloud stack.
A Claude-powered offensive-security suite — 23 attack modules, AI false-positive verification, and a live dashboard. Used internally to harden the portfolio and in client security engagements.
Humanify
Because robots don't buy.
Ad-fraud and bot detection for marketing agencies — score every account, block bot clusters in real time, and prove to clients exactly how much budget you protected.
AI Operations
Act at scale — the execution layer for marketing and infrastructure.
Bertha — Agency OS
An entire agency, as one operating system.
An enterprise Agency OS, three years in the making on a patent-filed stack. 19 microservices run the whole marketing lifecycle — from CRM mining and strategy to multi-LLM creative, omnichannel placement and feedback — with enterprise-grade security, governance and a smart LLM routing layer underneath.
K8 Inspector
Stop fighting Kubernetes. Start mastering it.
AI-powered Kubernetes operations — natural-language queries, predictive issue detection, integrated security scanning and one-click remediation across every cluster you run.
LinkedIn Poster
Thought leadership, in your voice — on autopilot.
AI writes LinkedIn posts in your voice for your audience, designs the visuals, picks the best time to publish and grows your network — with a human-approval gate before anything goes live.
BlackBoxAI
One human. One hundred robots. One action at a time.
Free and open-source. The AI operations manager for robot fleets — it doesn't show you a hundred green and red dots, it decides which robots need a human, why, and what to do next. The robotics endpoint of the Supervisor Pattern — one supervisor, many autonomous workers.
Bertha Featherweight
The whole agency, minus the infrastructure.
Bertha's judgement without Bertha's microservices — the Agency OS distilled into a single Claude Code skill. Seventeen reference engines and one operating doctrine, installed in the time it takes to make a cup of tea.
SEOboss
Audit, fix, prove — for search engines and answer engines alike.
A local-first SEO and answer-engine platform that does not stop at the audit. It crawls, runs 41 evidence-backed checks, drafts the actual fix, applies it through WordPress, Shopify or Cloudflare behind an approval gate, and then proves on the next scan that the fix moved the number.
Human Enablement
Remove the friction between human intent and finished work.
SpeakEasy
Your voice. Perfectly written.
A macOS app that turns natural speech into polished, professional text anywhere on your Mac — filler removed, grammar fixed, punctuation added — with a coding mode for developers.
CodeEasy
The command center for the age of AI co-pilots.
A visual command center for AI-assisted development — see your codebase seven ways, watch Claude Code make changes live, and orchestrate autonomous multi-agent builds with human-in-the-loop gates.
CloudCode
Ship code with AI — not your codebase.
A private-by-default coding agent that runs on the models you control — on your own machine or inside your own cloud. Built for work that's under NDA, under regulation, or simply too valuable to send to someone else's API.
Bertha Frontend Tester
Test every journey before your users do.
AI-driven frontend QA for React apps — auto-generate user journeys from your pages and source, record and replay real flows, catch bugs, and log them straight to Jira from one dashboard.
Saksi
Conversation becomes executable memory.
Formerly Sidekick. An AI personal assistant that listens to your working life and turns it into trustworthy, queryable, long-horizon memory — memory precise enough to act on, and it never acts without your say-so. Runs privately, hosts anywhere, forgets nothing.
Platform
The connective tissue the portfolio stands on.
License Server
The licensing and distribution backbone.
A Lemon Squeezy-style license server — key management, machine fingerprinting, Stripe billing, S3 software distribution and a customer portal — quietly powering the rest of the portfolio.
ATLAS
Agents you can afford to let run.
The product name for AOS — the agentic operating system, grown out of The Governor. A self-hosted agentic AI service in miniature that bounds what an agent may spend, proves whether it did the work, and undoes it when it didn't.
Moonshot
A societal bet, not a workflow tool.
Educational · working builds
Not for sale — real, working tools I build to think in public and teach the architecture. Each one links straight to the long-form write-up that explains it.