Sample Work

Working systems, not strategy decks.

These demos and founder-built products show how we turn messy operations into tools people can actually use. Each one explains the problem, who it is for, where people stay in control, and what is real versus simulated.

Public marketing site homepage. The matching workflow is not shown.

Pilot in development13 Canons venture | Commercial-building intelligence

RetrofitSignal

Source-cited matching against public records

A founder-led pilot that turns public commercial-building records into source-cited retrofit opportunity matches, with AI-assisted analysis and manual verification before any introduction.

Problem: Buildings that may qualify for energy-retrofit incentives can only be found today by cross-referencing scattered public records by hand.
Intended user: Teams that sell or deliver commercial-building energy retrofits.
Workflow: Pull public building and permit records, apply matching criteria, summarize each candidate with citations to its source records, then verify it by hand.
Human control: A person verifies every candidate match against its sources before any introduction is made.
What is built: The public marketing site shown here. The matching and verification workflow is in active development and is not shown.
What is simulated: Nothing simulated is shown. No product interface, customer data, or results are presented.
What we would measure: The share of candidate matches that pass manual verification, and the time it takes to verify one.

MVP completeStudio build | Program operations

MissionThread AI

Human approval before changes

An auditable program-operations MVP connecting requirements, schedules, budgets, risks, tests, suppliers, field feedback, deterministic analysis, and bounded AI support.

Problem: When a supplier slips, nobody can quickly see every requirement, milestone, and cost it touches across separate tools.
Intended user: Program managers and reviewers on complex hardware-and-software programs.
Workflow: Record a program event, calculate schedule and budget exposure in code, draft a plain-language summary and mitigation options, then send them for human review.
Human control: Mitigation options wait for a person to approve them. Reviewers get read-only access, and every analysis run is kept in an append-only history.
What is built: A working MVP: connected program records, deterministic exposure calculations, AI-drafted summaries that cite their source records, role-based access, and audit history.
What is simulated: The program, its records, and its events are fictional, synthetic, unclassified demo data.
What we would measure: Time from a recorded event to a reviewed impact summary, and how often a drafted summary needs correction.

CompleteStudio build | Experiment operations

Signal Dash

Documented statistical assumptions and limitations

A lightweight ExperimentOps MVP covering intake, workflow status, exposure and conversion tracking, two-proportion analysis, decision records, and optional AI summaries with deterministic fallback behavior.

Problem: Experiment results end up scattered across spreadsheets and chat, with no clear record of who decided what and why.
Intended user: Small product and marketing teams running A/B tests.
Workflow: Register an experiment, record exposures and conversions for each variant, generate a statistical readout, then record a ship, hold, or kill decision.
Human control: A person records every decision next to the readout it was based on. The AI summary is optional, and a deterministic summary replaces it if that step fails.
What is built: A complete MVP: experiment intake, per-variant tracking, two-proportion z-test readouts with documented assumptions, and decision records.
What is simulated: The experiments in this screenshot are sample data, not a client's results.
What we would measure: Time from an experiment ending to a recorded decision, and the share of decisions linked to a readout.

These are studio-built products and prototypes, not client performance claims.