What Got Built
AI platforms, evaluation engines, governed legal systems, voice pipelines, workflow automation and 190 visual explainers — six domains, all deployed to production.
AESOP Transformation OS
Full-lifecycle agent development platform — discover, design, build, evaluate, repair, track. The evaluation engine that later powered Legal-OS’s Harvey monitoring layer.
The hardest problem in AI: how do you evaluate an agent when the agent cannot be trusted to self-assess? Twenty-plus specialised evaluators each score one dimension, then a programmatic engine applies weighted tiers, veto rules and certification levels. No LLM self-assessment.
50 agents · 75 API routes · 76 migrations · ~990 tests
FastAPI + Supabase + Celery · Next.js 16 · Fly.io + Vercel
Red Team, Bias Probe, Sounds Like AI. Hard veto at 75.
Legal AI Operating System
A governed platform for building, deploying, measuring and independently evaluating AI across the legal enterprise.
Five layers — knowledge, governance, functions, operations, org model. The breakthrough is independent Harvey evaluation, ported from AESOP and adapted for legal outputs. Ten functions, including a firm simulation: a Monte-Carlo digital twin that runs your P&L under different AI levers. Legal-engineer verification cite-checks every filing against real case law. Above the ten sits the Fault-Line Radar — next section.
10 functions · ~104 endpoints · 120 tests · 50 migrations
RLS client isolation · ABA 512 ready · legal.sickofancy.ai
Harvey can’t grade its own homework. Legal-OS does.
Fault-Line Radar
A prediction engine for legal AI. It forecasts where the next ruling or required control lands — and it grades its own calls against what actually happened, then shows the math.
Eleven fault lines, each scored 0–10 across five meters: the capability, the ruling, firm adoption, vendor momentum, and market supply. Evidence counts by the authority of its source, not its volume — one ruling outweighs ten vendor posts. Every ninety days the engine replays using only prior evidence to check whether it flagged a fault line before it broke. Above the forecast sits the advisory answer: govern it, or deploy it.
11 fault lines · 5 meters · 76 dated evidence items · 6 streams
authority-weighted scoring · self-grading 90-day replay · weekly regeneration
It won’t give you the date. It grades its own calls, and the math stays on the page.
Strategic simulations
Two multi-agent simulations that turn “what does AI transformation actually look like” into concrete, repeatable, falsifiable answers — one about the economics, one about the change process.
An AmLaw 100 firm adopts AI across sixteen sprints and pricing decides everything: hourly billing makes AI a net loss, alternative fee arrangements make the same AI a net win. Same tool, same people, same adoption. Alongside it, two identical copies of one insurer run through organic discovery versus traditional change management — same hurricanes, same attrition, only the change process varies.
16 sprints · two experiments · deterministic engines · seeds 42 + 7
Design drives adoption. Discovery drives outcomes.
SuperAssistant, TSE & TBG Automations
Voice interview pipeline, email automation and workflow orchestration for a consulting firm — serverless across Railway, Supabase and ElevenLabs.
Multi-tenant agent orchestration with structured output pipelines and vector-search knowledge bases. Serverless voice interviews with branching question flow, real-time transcription and analysis. Plus the operational layer: cohort automation, Gmail triage, ecosystem mapping, and Walter — a fully client-owned assistant.
19+ API route modules · Railway + Supabase + Cloudflare · CI/CD
Manual workflows, turned into pipelines.
190 visual explainers
Self-contained HTML pages — architecture diagrams, maturity models, process maps, audit reports, platform showcases — spanning every project.
32 AESOP · 42 Legal & radar · 6-stage Monday pipeline · TBG audits
No frameworks. No build step. Renders instantly.