A public healthcare data platform needed a steady supply of trustworthy insights, publication-quality visuals, and constant quality control. Our principal led the platform's delivery and personally designed and built the three AI systems below. They were designed as one system, not three tools: what one produces, the next one uses. Identifying details are withheld; everything else is as-built.
A public data platform goes stale without a constant stream of new insights. Producing them normally means analysts spending hours a week scanning news, journals, and datasets for material, and that scanning is the first thing dropped when real work gets busy.
Tessera is an autonomous agent that scans the public web every day for emerging topics in healthcare and aging, filters them against the platform's editorial priorities, and drafts structured insight candidates: claim, evidence, sources, and a suggested visual. Each candidate comes out as structured insight JSON, the same object Emblema reads to build its visuals. A human editor reviews and promotes; the agent never publishes on its own.
A daily pipeline of vetted insight candidates feeding the platform's content calendar, with batch runs producing 15 publishable insights at a time. In its first seven weeks of operation, daily scans have captured more than 150 structured signals, every one carrying its provenance trail. The agent supports the process; every greenlight is a human call.
autonomous scans, on schedule, unattended
insights produced in a single batch run
human-reviewed before anything publishes
Every insight worth publishing needs a visual, and every visual used to need a designer. The design queue became the bottleneck between "we know something" and "the public can see it."
Emblema takes a written insight and produces a branded, publication-ready infographic: correct typography, correct palette, correct data presentation, exportable for web and print. Give it a Tessera insight's JSON and it drafts the infographic on demand, or it drafts on its own schedule without anyone triggering it. It's also self-serve, so the person who found the insight can produce the visual by hand, and the brand system is enforced by the tool rather than by review cycles. The same insight objects can feed a separate content-generation engine, so one finding becomes both a visual and drafted copy. The five-color data palette was validated colorblind-safe before launch, and eight named styles cover the common insight shapes.
The design bottleneck is gone from the publishing path. Visual output scales with the insight pipeline instead of with headcount, and every graphic that leaves the tool is already on-brand.
design-queue wait between insight and visual
infographic styles at launch
analysts produce finished graphics themselves
A public platform spanning dozens of data products, articles, and interactive pages can break quietly: a data refresh shifts an axis, a link dies, a number stops matching its source. Manual spot-checking catches a fraction of it, late.
Sentinel audits the live platform against two other views of the truth: the production codebase and the canonical published data. Any disagreement becomes a tracked finding. It runs 46+ distinct checks across 8 suites, roughly 384 check instances per sweep, every weekday morning, and writes five audience-tailored reports: developers, data team, communications, design, and leadership. It's the quality gate for everything that reaches the platform: insights Tessera surfaces and visuals Emblema renders still have to clear Sentinel's checks once they're live, the same as every other page and number the platform publishes.
Defects surface in hours instead of whenever someone happens to click the right page. In its first sweeps it caught a public page returning a server error and a broken dollar figure inside a live calculator. Leadership gets a trustworthy quality signal; engineers get specific, reproducible tickets.
automated checks in continuous operation
audit suites, from data accuracy to accessibility
audience-tailored reports from every run
Each of the three was built to be the next one's input. That is the difference between three tools and one system.
Scans the public web every day and writes each finding as a structured object: claim, evidence, sources, suggested visual.
Reads that same object and renders the on-brand infographic, on request or on its own schedule. No re-entry, no design queue.
A person reviews the insight and its visual together and decides what goes live. Nothing publishes on its own.
Tessera's output is a structured object rather than a document, so Emblema reads it without anyone retyping the finding. Emblema's visuals land on the same platform Sentinel already audits, so they're checked like everything else. Adding a fourth piece means plugging into what's there, not starting over. Client work gets the same standard: every build should make the next one cheaper.
Tell us which workflow hurts. We'll tell you what the version for your organization looks like, what it costs, and how long it takes.