Five concurrent investigations into applied AI.

Kronos Engine forecasts crypto price as a calibrated probability distribution and uses that distribution to discipline how capital is sized — it de-risks rather than predicts. It's backtested and deploy-ready, but not yet live-traded. The honest reckoning came from out-of-sample testing on a 74-day, 1,800-candle XRP window: the model has no directional edge (~48.6% hit rate, a coin flip) and lost to a naïve persistence baseline on volatility by ~38%, so I cut both of those uses entirely. The one property that held up was calibration — about 79% of realised prices landed inside the 80% forecast band — so the engine now leans on that single verified strength. It sizes off how wide the distribution is, and steps aside ahead of scheduled macro events.

Most people's personal data is scattered across fifty-plus data brokers and people-search sites that scrape, package, and sell it. Manual removal is hours of repetitive work per broker, and the records reappear. Privacy Posture applies an AI agent — operating with human approval gates — to map the PII footprint, draft jurisdiction-correct removal requests, track every request through to confirmation, and re-check for reappearance on a schedule. Effectively a small Security Operations Centre where the analyst is an LLM, the case management system is a structured workspace, and the legal-basis library is encoded as a template set.

A research-aide toolkit for solo bug bounty work, first deployed on HackerOne. Architecture is intentionally split: standard scanners (nuclei, subfinder, httpx) handle the technical execution, while a multi-agent Claude layer — planner, researcher, and report-drafter — handles scope parsing, scanner orchestration, and turning collected output plus researcher observations into platform-formatted draft reports for human review. The most distinctive capability is the report drafter: turning structured scanner output and unstructured notes into submissions ready for human approval.

Before I pull an unfamiliar repo onto my machine, I want a fast read on two separate questions: is this code likely to harm my system, and is the project actually maintained or quietly abandoned. You paste a public repo URL and it pulls the repo's GitHub metadata, scans the source for known risky patterns, checks for known vulnerabilities against the OSV database, and runs a semantic check on whether the repo fits what I said I wanted it for. It returns a single-page dossier that deliberately splits a Risk Score ("will this code hurt me") from a maintenance Hygiene Score ("does the maintainer take security seriously"). Every run is appended to a Google Sheet, so I keep a running history of everything I've checked.

Every time I started a new chat with an AI, I re-explained who I am, what I'm building, and what we'd already decided — and the moment the conversation ended, all of that was gone. The thesis is that a second brain shouldn't be a passive archive I search through; it should be the AI's working memory, structured so an agent can read it on demand and write back to it. So I built a Drive-based vault organised by entity type (projects, people, tools, concepts, decisions) where every note opens with consistent metadata, an operating manual tells Claude how to navigate it, and session rituals make sure nothing useful disappears between conversations. I parsed my own full chat-export history into it, then synthesised the raw material into linked wiki pages. It now runs as the context layer behind how I actually work day to day.