Proof, not promises.

A live, browser-based voice simulator that puts you on a real-time phone call with an AI person in crisis — 911 dispatch, hostage negotiation, undercover sting.

↳ A live, publicly playable real-time voice AI product deployed to production with four distinct scenarios.

AI that calls dental insurance payers, navigates the IVR, verifies benefits, and gets the eligibility PDF faxed back — giving front-desk staff back 15+ hours a week on hold.

↳ Validated end-to-end against a real national payer's production IVR — the AI navigates the full menu tree, verifies eligibility, and delivers the fax-back PDF in ~3-5 minutes, freeing an estimated 15+ front-desk hours per week. Currently piloting with practices.

Browser and telephony voice agents that answer and place calls, pitch and qualify leads, handle objections, and book or route the caller — deployed for sales and front-desk reception.

↳ A reusable playbook for sales and reception voice agents spanning browser and telephony, provider-agnostic across the major real-time voice stacks and grounded to a client's real products, pricing rules, and booking systems.

Conversational voice self-service for banking and credit unions — members speak naturally instead of pressing 1, and the agent completes real transactions against the core banking system.

↳ Working conversational voice self-service demos that authenticate callers and complete real banking transactions against core-banking APIs, positioned as a natural-language replacement for legacy menu-tree IVR.

Ingests customer interactions across channels into a queryable knowledge graph, exposed through a multi-tenant dashboard and API.

AI-scored quality analytics for contact-center interactions — real-time call coaching and automated, SOP-governed QA evaluation with human calibration.

↳ A quality-management stack that scores interactions at volume instead of by sample — real-time on live calls and automatically after the fact — while keeping humans in the loop through calibration, override, and full scorecard-to-coaching traceability.

AI that scores every mortgage sales call in real time — classifying it against an 8-stage loan-officer funnel and an 84-point diagnostic taxonomy — to flag compliance risk and surface coaching before the deal is lost.

↳ A live, self-serve product that scores mortgage sales calls at volume instead of by sample — funnel stage, 84-point diagnostics, and compliance auto-fails — with every score traceable to the moment in the call that caused it.

Upload a daytime house photo, get a professional nighttime lighting render.

Self-serve SaaS that automates open-source asset and identity research on judgment debtors.

AI-native case assembly for family-based U.S. immigration filings.

↳ A V0 platform whose spine is the typed case record: LLM steps are constrained to classification and cited extraction, OCR and model calls sit behind swappable interfaces, and a regression eval guards quality on every change.

A marketplace where customers post jobs and local providers compete on price and quality.

An in-car robotic co-pilot that physically operates your phone on your behalf — voice in, robotic stylus arms out — so drivers can text, navigate, and use any app hands-free without touching the screen.

↳ A concept-and-IP-stage project: patent filed, industrial-design specification and renders locked, full system architecture documented, and an early-interest landing page shipped ahead of hardware.

A wheeled rover that prints sidewalk-chalk art on driveways — pick a template in an app, hit go, and it drives a planned path spraying liquid chalk through a multi-color print head.

↳ v0.1 software pipeline works end-to-end: the example road-grid design flows cleanly through planner and simulator (3/3 unit tests pass). Hardware is still ahead of the software — firmware parses toolpaths but does not yet drive motors or spray.

A 'whose turn is it?' companion device for UNO that uses computer vision to recognize played cards, backed by a Flutter app and ESP32 firmware.

↳ The CV pipeline ran end-to-end and hit 92.6% on 54 synthetic fixtures — clearing the 90%+ feasibility gate — with the report explicitly cautioning that the real go/no-go must be re-run on physical webcam captures, not template-derived synthetic images.