The Lab
Advising on AI transformation means making promises on other people's budgets. So before a pattern goes in front of a client, it goes through my lab first: production systems, real users, real cost ceilings, real failure modes. Everything below ran past the demo stage. Some of it fails on a schedule, gets caught, and fixes itself. That's the point.
Case studies
An AI investment research system over 250 Nifty-listed companies: evidence pipeline, multi-factor scoring, and a paper portfolio whose math refuses to flatter. The day it caught my own dashboard counting deposits as gains is the day it graduated.
Running dailyAn AI SDR that works on top of your CRM: the research, the LinkedIn signals, the drafts, the morning brief. Governed by a constitution, capped by budgets in code, vetoed by a human. The agentic CRM blueprint, given away in full.
Running dailyA voice-first execution engine. Ten seconds of speech becomes your tasks, your people, and a morning brief. Audio never leaves the phone, and your voice never trains anyone's model.
Private betaRelationship intelligence for people who sell: a signal graph over your professional network, warm-path inference to the people who matter, and outreach recommendations you approve.
LiveShared infrastructure
A single LiteLLM gateway fronting three production AI systems: model groups named by the job (draft, classify, judge), automatic failover chains, and a self-running 'Gateway Doctor' that probes, quality-scores, promotes and demotes models daily. I never quote a client a model cost I haven't paid myself.
What the lab is for
Every architecture I recommend has a running instance somewhere in these pages. A system with logs, not a slide.
The lab's job is to find the failure modes cheaply, in my systems, with my name on them. Client programs meet them second, prepared.
A lab project graduates when its pattern is transferable: governance that would survive an audit, math that would survive a CFO, privacy that would survive a regulator.
Every system in this lab was architected and built by AI agents under my direction. That's the same operating model I now set up for client teams. It's not a footnote, it's the proof.
Also from the lab
A hands-on exploration of agents, operating systems, and what changes when software moves from passive tools to active roles.
Open link ↗Public noteAn OpenClaw-style agent experiment for turning a repetitive personal workflow into a clearer, reusable operating pattern.
Open link ↗