Brian Zeichick

AI implementation for organizations where being wrong is expensive.

I'm a product designer and product manager with more than twenty years in healthcare technology. I help teams work out which parts of their work AI should touch, design the interfaces and guardrails around it, and ship the result — including the parts where the honest answer is to leave the process alone.

Currently building

ColdLade

In development with US Perishables

ColdLade brings AI to cold-chain freight operations. Perishable shipments arrive as emails, PDFs, scanned paperwork, and phone calls, and they change constantly. ColdLade turns that stream into accurate, current loads inside the systems carriers and brokers already use.

It catches the details that turn into expensive mistakes — temperature requirements, appointment conflicts, missing information — and keeps a person in the loop on every decision that matters. Less manual data entry, fewer missed details, and more of a dispatcher's day spent moving freight.

It's also the clearest statement of how I think AI should be built: narrow scope, a real operational problem, and a human holding the pen on anything consequential.

  • Document & email extraction
  • Human-in-the-loop review
  • Exception flagging
  • Logistics systems integration

What I do

Selected work

AI workflow rollout across a working team

Baszucki Group · 2026

Designed and staged the deployment of AI assistance across a communications and program team's existing tools — project tracking, documentation, chat, meetings, and file storage. The deliverable was an implementation matrix of roughly two dozen workflows, each specified in enough detail that a security reviewer could sign off on it before anything was connected.

Every workflow carried the same four decisions:

  • Read-only or edit access, scoped per connected system rather than granted wholesale.
  • Scheduled or human-triggered — chosen deliberately, because anything running unattended needs a far higher bar than something a person invokes and reviews.
  • An explicit approval gate before any action that writes, sends, or creates.
  • A named beneficiary and a priority tier, so rollout order reflected who was actually waiting on it.

A parallel matrix tracked everything deliberately not built: requests routed to tools the organization already owned, workflows left manual because the automation case didn't hold up, and items parked pending a platform decision. That list did as much for adoption as the build list — it made the rollout legible to people who were reasonably wary of it.

  • Connector permission modeling
  • Human-in-the-loop design
  • Scheduled vs. interactive
  • Phased rollout
  • Stakeholder mapping

Care coordination platform design

ThoroughCare · 2026

Product and interface design work with ThoroughCare, a care coordination platform whose programs span chronic care management, transitional care, behavioral health integration, remote patient monitoring, and annual wellness visits.

Care coordination software is unusually hard to design well because the same screen serves people whose jobs barely overlap. A care coordinator working a daily patient list, an enrollment coordinator working eligibility, and a chief medical officer reading performance across sites all need the product to open somewhere different. My work centered on designing for that spread rather than averaging across it — building role-specific views as interactive, click-through surfaces the people who do the work could use and react to directly, instead of static comps reviewed in a meeting.

  • Role-adaptive UI
  • Interactive prototyping
  • Component library
  • Care coordination workflows
  • Value-based care

Platform redesign and AI chatbot design

Viaduct · UX consultant

A redesign of Viaduct's legacy vehicle-data analytics platform, plus design work on its AI chatbot. I started by interviewing internal platform users to compare the workflow the software assumed against the one people actually performed. The gap was the finding: root cause analysis meant bouncing between several pages, which made the platform both slow to use and hard to learn.

On the chatbot — which summarizes vehicle claims and suggests how best to split the data for analysis — my work was on the conversational surface rather than the model: making the assistant's utterances plain, and seeding prompts that show users what it's actually capable of. An assistant nobody knows how to address is indistinguishable from one that doesn't work.

  • Conversational UI
  • Prompt scaffolding
  • Information architecture
  • User research
  • Design system

New platform and public benefits-rate tools

Paradigm Senior Services · 9 months

Built a platform from scratch for a senior-services company: design system, personas, and requirements, with the first shipped surface being a set of public tools that let home care agencies look up location-specific Medicaid and VA rates and programs.

Client interviews early on surfaced the real problem — the information people needed most was buried inside an account management system they found hard to navigate. That reframed the roadmap around surfacing it instead of reorganizing it.

I assumed desktop-first. Roughly 40% of clients and prospects were reaching the rate tools from a phone.

That finding forced a mid-project scramble to get mobile designs ready, and it's the reason I no longer take a usage assumption into a project without checking it against data. The design system was built to ADA-compliant contrast standards throughout, on Bootstrap 5.2, and deliberately close to marketing's existing palette so the two surfaces didn't drift apart.

  • Design system
  • ADA contrast compliance
  • Persona development
  • Public-facing tools
  • Medicaid & VA benefits

Earlier: product consolidation across three products at Home Care Pulse; dashboard, chart, and navigation standards at MedeAnalytics; and user research at ePocrates on advertising, secure messaging, and contact directories. Full case studies live on my portfolio.

How I work

Organize the day around what has to happen next — not around how many minutes have accumulated.

A few principles that show up in every engagement, regardless of domain:

Background

Twenty-plus years in healthcare technology, spanning product design and product management. I'm currently Senior Manager of Experience & Web Operations at Baszucki Group, a philanthropy funding work in metabolic and mental health, where I also lead Metabolic Mind's web presence.

Much of that work has been in benefits and regulatory territory — including public tools that helped people find what Medicaid and VA programs actually pay in their state. Getting an answer wrong there isn't an inconvenience; it changes what care someone believes they can afford. That's the experience shaping how I approach AI work now: the guardrails are the product.

Not all of it is healthcare. I've done platform and conversational-AI design in vehicle data analytics, and I'm building AI tooling for cold-chain logistics now. The method carries over intact — high-stakes operations, expert users, and an assistant that has to be honest about what it doesn't know.

I studied Mathematics and Art at UC Santa Barbara. I live in Oakland.

Elsewhere

Contact

brian@zeichick.ai

Currently taking on a limited number of engagements. Happy to talk through a problem before there's a scope.