About
Hannah Kraulik Pagade
AI product leader · Product management & UX design
I own strategy and launch for AI products in high-stakes domains: what to ship first, what success means, and how users should trust model output. My proof is live software—clinical intake, patient literacy, regulated-style document intelligence, enterprise-style conversational channels, and agentic infrastructure (Ask Hannah MCP: a live HTTP MCP server on Railway, registered in Claude.ai, with ten structured tools)—not decks alone.
On product management, I frame scope, priorities, and tradeoffs with engineering. On design, I define flows, empty and failure states, and how the AI behaves in the UI. On solo or early work I hold both; on a team I partner and flex where the gap is largest.
I'm a Licensed Practical Nurse at PAM Health Rehabilitation Hospital of Westminster, Colorado, and founder of Rohimaya Health AI. I prototype and ship in the stack when the team is small—Next.js, Vercel, Claude API—so product decisions are tested against real documents, transcripts, and utterances, not theory alone. MS in Artificial Intelligence and Machine Learning at CU Boulder, expected 2026 (in progress).
Outside tech: co-founded Two Peaks Chai Co. with my spouse—artisan chai in Westminster.
At a glance
- Five live AI surfaces on the work page—clinical triage, patient literacy, enterprise conversation lab, financial document intelligence, and Ask Hannah MCP (live MCP on Railway, registered in Claude.ai). Onboarding Agent is in build (coming soon).
- 17 years in healthcare operations leadership—staff scale, audits, cost outcomes—before shipping software; domain judgment carries into product and UX calls.
- MS AI/ML at CU Boulder (in progress, expected 2026); founder of Rohimaya Health AI.
- Hiring target: AI Product Manager and/or AI product design roles; open to relocation. Full case studies on this site.
How I work
- I start from constraints, not feature lists. Safety and governance, who the product is for, what “good” output looks like, and when the model must stay assistive instead of authoritative. The constraint set is the spec, whether the domain is clinical, financial, or enterprise conversational.
- I design the intelligence layer first. Structured outputs, attribution language, and guardrails in the prompts before the UI does the selling. If the model behavior is not trustworthy, the interface cannot fix it.
- I ship in small loops. I build in Cursor, deploy to Vercel or Railway, and test on real documents, transcripts, and utterances. Then I iterate on the failure modes users would actually hit in production, not just demo paths.
- I treat host integrations as the spec. When the distribution surface is a connector (for example Claude.ai and MCP), the platform integration requirements win over the quickest SDK path. That is how stdio-only prototypes become HTTP MCP, Zod-guarded tool inputs, and deployments that still behave after redeploy.
- I judge success by clarity of scope. Someone using the product should know what it does, what it refuses to do, and what to do next, without mistaking an assistive tool for an authority they should not rely on alone.
- I collaborate like a product partner. Specs, acceptance criteria, and review of model behavior with engineering; design feedback that accounts for latency, cost, and failure modes—not only ideal paths.
What I built and shipped
Clinical intake and triage: structured assessment, red-flag surfacing, urgency tiers, history, and billing. Early commercial pilot with subscriptions, credit packs, and server-side usage enforcement.
Product: Defined paid vs. free tiers, credit packs, and enforcement that moved server-side after real per-session cost exposure—not pricing on slides only.
UX & AI behavior: Meridian Oracle system across funnel, errors, and email so the product reads as clinical authority without feeling cold; typed assessment output reviewers can audit, not an unstructured chat log.
Next.js 16 · TypeScript · Tailwind v4 · Claude API (Sonnet/Haiku) · Supabase · Stripe · Resend · Vercel
Free, no-login translation of clinical documents into plain language—twelve languages, Simple / Clear / Complete reading levels, urgent items first, user-initiated verification, 90-day shareable sessions.
Product: Shipped user-initiated verification (Check for Missing Info) with issue cards instead of silently re-running every translation—traded always-on latency and cost for patient control and trust.
UX & AI behavior: Plain reading-level labels (not grade numbers) and THOROUGH CHECK / PARTIAL CHECK / QUICK CHECK badges so verification is legible; side-by-side and accessibility patterns as launch requirements.
Next.js 15 · Claude Sonnet · Zod (API requests) · Supabase (90-day) · Web Speech API · Vercel
Conversational lab: one utterance drives IVR, chatbot, and agent assist in parallel with NLU surfaced; SSE streaming, five sentiment themes, Whisper transcription, OpenAI Realtime Live Call.
Product: Empty state plus Live Call as primary paths so the story reads on mobile; SSE so parallel channels unfold in real time instead of one blocking JSON blob.
UX & AI behavior: data-sentiment retints the full chrome—bereavement and urgency are environmental, as on the bereavement path—not a badge bolted on after layout.
Next.js 16 · Claude (SSE) · OpenAI Whisper · TTS · Realtime · Vercel
Six-section structured intelligence for earnings calls and filings—not a summary—plus compare-with-accordion, branded PDF/PPTX, 30-day Supabase deck URLs. Assistive only, not financial advice.
Product: When Canva Connect was blocked, owned the deck layer: Claude outline, pptxgenjs, pdf-lib, and a custom /deck viewer—removed dependency on a vendor gate.
UX & AI behavior: Source anchors prompt-pushed with schema-soft validation; compare opens highest-signal deltas first so users scan before they deep-read.
Next.js 16 · React 19 · Claude Sonnet · Zod · Supabase · pdf-lib · pptxgenjs · Vercel
Live Model Context Protocol server on Railway: ten tools expose profile, projects, metrics, skills, voice, FAQ, a recruiter hiring brief, and resume and cover letter generation from structured data. Registered as a public connector in Claude.ai so hiring managers can query Hannah's professional facts inside the tools they already use.
Product: Scoped for connector reality (HTTP streamable MCP, not stdio-only), strict generation contracts, and conversion (hiring brief contact ordering, UTM tagging)—a portfolio surface that still behaves like a product.
UX & AI behavior: The UX of an MCP tool is its schema: field shapes and first-person voice data are designed so downstream synthesis reads correctly to humans, not like a press release.
Node.js 20 · TypeScript · MCP SDK · Express · Zod · Anthropic API · Railway
All code at github.com/rohimayaventures
My background
I have 17 years of healthcare operations leadership across acute care, post-acute rehabilitation, and senior living. Real employers include Amberwood Post Acute Rehabilitation, Center at Northridge, and a senior living facility in the Westminster area. I managed 130+ staff, achieved $1.2M in cost savings through workflow optimization, and maintained 96% regulatory audit success rates.
That work trained the muscles product teams need: operating under regulation, aligning nursing and leadership under pressure, communicating risk clearly, and improving workflows when the downside of a bad decision is real—not a postmortem slide.
I did not come from a product whiteboard. I came from the floor. Those 17 years were field research.
What I'm looking for
I'm focused on AI Product Manager and senior AI product design roles (product designer or UX designer owning AI-native flows). I also consider founding PM and Head of Product where scope includes both strategy and hands-on AI UX. Open to relocation and hybrid. If this sounds like a fit, use the contact form or connect on LinkedIn.
Product & strategy
AI & product design
Core technical stack
Execution leverage with engineering—not a backend specialist. Depth varies by product; case studies list the exact bindings.
Python · FastAPI · Cloudflare · ElevenLabs · pdf-lib · pptxgenjs · Cursor · Node.js · Express · MCP SDK · Railway — see case studies for product-specific stacks.
Get in touch
I'm interested in conversations about AI product management, AI product design, and founding-team roles where both strategy and shipped UX matter. Recruiters: share the stack, stage, and how AI shows up in the roadmap.
Use the contact form to reach me, or connect on LinkedIn. Explore the live products: OrixLink AI, HealthLiteracy AI, ClearChannel by Vestara, FinanceLens AI, and Ask Hannah MCP.
Resume available on request.