Shipped, in production.

ViewSonic's catalog runs to 1,618 products across 55 retailers. Watching revenue, competitive rank, and pricing compliance over a surface that size by hand is a job that never finishes.

Ten AI agents run continuously against the catalog, each owning one narrow slice of the work. Their output lands in a single multi-tenant command center rather than ten separate reports.

$31M of year-to-date revenue is tracked across the catalog, the brand ranks first against nine competitors, and pricing violations across 55 retailers are flagged automatically instead of found by hand.

A $40K ticket package only pays for itself if every event inside it is priced and sold deliberately. Priced event by event without a model, the value that is left unsold is invisible.

The system issues per-event pricing recommendations and attaches a confidence score to each one, so a recommendation can be taken or overridden on its own merits. Sellable inventory nobody had listed is surfaced in the same pass.

142.7% of the original $40K package cost recovered, measured against that cost basis, and $57K+ in additional sellable inventory identified automatically.

For a commercial security firm the best prospect is a business next door to a crime that just happened. That signal is public, but it arrives as scattered police incident feeds, not as a lead list.

The engine ingests live incident feeds, resolves every business within range of each commercial crime geospatially, then dedupes and scores what it finds. Territory coverage is drawn on a live map so a rep sees their own patch.

4,743 scored sales leads generated from live police incident data across 405 territories, and still accumulating as feeds update.

Campaign staff work from a statewide voter file, an FEC filing history, and a list of candidates — three datasets that never join cleanly, so the questions that span them get answered by hand or not at all.

A campaign OS built on a 23.0M-record statewide voter file joined against $2.365B of FEC money data, with donors matched across both sides and candidate scoring in the same view. Access is per-user, not public.

23.0M active voters tracked statewide, 245,954 FEC-matched donors representing $1.145B raised, and 174 candidates scored across California's 2026 campaign finance cycle.

A compounding pharmacy serves three audiences doing three different jobs — clinicians prescribing, patients following a protocol, admins reconciling the business — and every one of those paths touches protected health information.

Three distinct portals over one data model. Prescription routing, exception queues, protocol management, and revenue reporting each sit in the portal of the person who owns that work, rather than behind one shared admin screen.

All three portals run in production, covering routing, exceptions, protocols, and revenue reporting. No throughput figures are published for this build.

Biotech catalyst research is a filtering problem. Hundreds of candidates, a handful worth attention, and the facts that decide it scattered across filings and trial registries.

A research pipeline scores every candidate against a five-filter model and puts catalyst dates, modeled upside, and 13F ownership next to the score. Scrapers keep the inputs current; a private watchlist tracks pre-IPO names the public feeds miss.

Hundreds of candidates are scored on each run, with catalyst dates and ownership attached. The platform models upside; it does not track realised returns, and none are claimed here.

Auditing medical billing for compliance means checking claims against Medicare and Medicaid rules at a volume no reviewer can cover by reading.

A Medicare/Medicaid source-of-truth dataset behind an admin dashboard, with background workers running the audits and versioned migrations keeping the schema honest as the rule set grows.

The platform audits billing for compliance at scale and is live at servantx.ai. No audit-volume figures are published for this build.

People researching peptides end up on forum threads and vendor copy, with the reference material in one place and the arithmetic in another.

A per-compound directory of research pages plus an interactive dosing calculator, so the reference and the calculation sit on the same site.

The directory and calculator are live at thepeptidepedia.com. No traffic figures are published for this build.

Ford Raptor owners were spread across forums and groups, with shop recommendations, mod details, and trail knowledge living in comment threads nobody can search.

One hub: a hand-built shop map, a mod database, a build registry, and a trail directory, plus a truck-sponsorship program. The same codebase is wrapped for mobile rather than rebuilt.

60 shops mapped across 23 states and 35 Raptor specialists listed. Live at raptorcentral.com.

Financial guidance is usually prose, but the questions underneath it — will this retirement plan hold, what does this options position actually pay — are arithmetic, and arithmetic is easier to trust when you can move the inputs yourself.

16 simulators, one per question: Monte Carlo retirement, options P&L, tax scenarios, debt payoff. Live sliders on each, and every scenario produces a shareable link.

16 interactive financial simulators shipped and live at moneysims.com.

Two media brands, several writers, one SEO backlog. Coordinating briefs, rewrites, and updates across that becomes the work instead of the writing.

A shared workbench fronted by an AI editor that briefs new pieces, rewrites drafts, and updates published posts. Both brands run through it, separated by data rather than by tooling.

Multi-writer SEO operations for two brands run in one workbench with the AI editor in the loop. No output or ranking figures are published for this build.

An invite-only deal desk still drowns in decks. Reading each one down to a first-pass judgement is the bottleneck, and it is the same bottleneck every week.

Founders submit a deck; the platform scores and diligences it in seconds and matches it to investors. A conversational intake tool takes a pitch and returns a structured deal sheet, so the desk starts from fields rather than slides.

The deal desk runs invite-only and is live at defactoventures.com, with scoring, diligence, and investor matching in production. Deal volume is not published.

Staffing an international engineering bench out of Pristina is a positioning problem before it is a recruiting one: the bench has to look as credible as it is.

A recruiting agency site under an international-hiring umbrella brand, presenting the bench and taking enquiries directly.

Live at hirekosovo.com. Placement figures are not published for this build.

Learners stall mid-lesson, and the answer is usually one question away — but asking it means leaving the course to go and search.

A course platform with the AI tutor built into the material, answering in place while the learner keeps their position.

Live at truos.ai with the tutor in the learning flow. No enrollment figures are published for this build.