Executive Summary
GetBlue is building one intelligent companion that brings wearable data, lab results, and real clinical expertise together into a single personal experience. The vision is strong and the timing is right. The company also holds two assets that competitors cannot easily copy: founders with rare clinical, performance-data, and operating credibility, and a proprietary coaching engine built over eight months of focused work. To reach the August launch, that foundation now needs engineering built specifically for AI-native products.
This proposal comes with evidence. Over one 48-hour sprint, I designed, built, tested, and deployed a working GetBlue MVP: a member app with a conversational AI coach, live device connectors, a lab-report pipeline, and a separate clinical portal. It runs on a real backend today, covered by 130 automated tests and a production-grade release process. The demo is not the finished product; it is evidence of method and velocity.
The strategy: keep what is proven and make it exceptional. The existing engine stays intact, migrates into this hardened architecture, and gets supercharged with AI-native design — instrumented, testable, provider-neutral, and improving with every member interaction. Public beta in the first week of August. Continuous refinement through December.
WORKING MVP — HOURS
ADVERSARIAL SUITES
VIA JUNCTION
MEMBER + CLINICAL
CUSTOMERS ONBOARD
The Founders Are the Edge
Health companions are easy to launch and hard to trust. GetBlue starts with an advantage most competitors cannot buy: its founders. Dr. Conrad brings the authority of a career at the top of academic surgery. Dr. Wagner built Sparta Science into the standard for data-driven physical-readiness assessment across professional sports, the military, and health systems. Rand Currier co-founded Granite Telecommunications and scaled it into one of the largest privately held communications companies in the country, bringing the operating discipline to match.
That credibility lives inside the product. The coaching engine encodes the founders’ judgment, and members will trust GetBlue because its intelligence traces back to real clinicians and proven operators. Everything in this proposal is built to protect that asset and deliver it faithfully, at scale.
A Working Demonstration
To show method and velocity in concrete terms, I built and deployed a working GetBlue MVP in 48 hours: a member app and a care-team clinical portal on one secure backend. Everything below runs today — and you can try it yourself.
The Connector: one link for every device and lab
Members link Garmin, Oura, Whoop, Fitbit, Apple Health and more through Junction (junction.com), a dedicated health-data infrastructure platform with 300+ device integrations and 140+ healthcare organizations on board. The demo already ingests Apple HealthKit and Garmin data, backfills six months of history, and analyzes it overnight. Incoming data is normalized, deduplicated, and rate-limited, with tested handling for missing days, conflicting devices, and malformed payloads. Junction also supports direct lab ordering through LabCorp, Quest, and others across all 50 states, with digital results flowing into the member record and to the clinical team.
Blue, the AI coach
Conversational by text or hands-free voice. Every recommendation arrives as a structured, reviewable proposal that applies with one tap and can be undone for 15 minutes. Members can challenge any suggestion: Blue explains its reasoning, adapts to what it learns, and escalates to the human clinical team when a question deserves a professional answer. Expert mode leads with data, mechanisms, and research; Casual mode is warm and human-centered.
An experiments engine
GetBlue proposes small, low-friction interventions and measures whether they work for that individual. Example: 10 mg melatonin and 100 mg magnesium before bed for one week. Daily check-ins plus wearable sleep data produce an honest verdict at week’s end. The same machinery extends to CGM: glucose data, logged food and drink, and tracked activity reveal each member’s personal glucose drivers.
A daily rhythm
A Morning Brief with a readiness score and a few achievable actions, a midday Pulse check-in, and an evening Ledger that seals the day. Small, Duolingo-style wins feed the next morning’s plan and build streaks of follow-through.
Labs and the clinical portal
A member photographs a lab report. The system parses markers, ranges, and trends, then holds results as “awaiting review” until a clinician signs off. In the portal, the care team reviews member data and labs, proposes plan changes for the member to approve, manages coaching agendas and secure messaging, and can arrange additional non-diagnostic tests.
Try it live
Both applications are deployed and available now. Explore them the way a member and a clinician would.
GetBlue trial app
The full member experience: AI coach, daily rhythm, wearables, labs, and experiments. Create an account or explore the demo sandbox.
getblue-v3.metadata.cc ↗Clinical portal
The care team’s view: member roster, lab sign-off, plan proposals, and secure messaging.
getblue-admin.metadata.cc ↗Trial environment with demo data. Please treat access credentials as confidential.
FULL ACTION PARITY
SURFACES, ONE MAP
AUDITED PER RELEASE
TYPES FROM THE COACH
BACKFILLED & ANALYZED
AI-Native Engineering Is the Whole Game
An AI health product carries higher stakes than ordinary software. A failure here means a wrong suggestion about a supplement dose, a leaked health record, or a fabricated lab interpretation. Trust is earned by the engineering system around the model. The demonstration was built under a written standard for exactly this, and every rule below is enforced in the running system today:
Permission parity
The AI can never do anything the signed-in user could not do themselves. Automated tests enforce this on every release.
Confirmation gates
Consequential actions such as finalizing a plan, importing labs, or deleting records always require explicit human approval, and applied changes can be undone.
Clinical authority is structural
Lab results require clinician sign-off before members see them as reviewed, and care-team proposals require member consent. These authority lines live in the data model itself.
Full audit trail, zero leakage
Every AI call is logged with cost, latency, and outcome, and every sensitive action leaves an audit event. Health data and personal information never enter the logs; diagnostic exports are sanitized automatically.
Adversarial testing, gated releases
The test suite deliberately attacks the system with prompt injection, permission bypass, abuse, and runaway-cost scenarios. 130 deterministic tests and a 32-screen visual audit run before any release.
Predictable AI spend
Per-user and per-feature rate limits and token budgets keep costs controlled and abuse contained.
This discipline is documented. I authored a formal engineering standard, Agentic Harness Engineering for AI-Native Production Apps, which governs how AI agents build, test, operate, and recover software safely. Quality and safety come from a repeatable system. That system allowed a safe, working platform to emerge in 48 hours, and it keeps one safe at production scale.
Launch speed without safety trade-offs. Every product claim can be verified by a test or a log. AI costs are budgeted and enforced per user. And because the system documents and audits itself continuously, a future engineering team can maintain and extend it with confidence, protecting the company’s investment well beyond this engagement.
Keep the Engine. Turbocharge It.
GetBlue’s most valuable technical asset is the engine: the clinical knowledge, coaching logic, process workflows, and model reasoning the founders have developed and refined over eight months. This proposal keeps all of it. The engine migrates intact into the hardened architecture demonstrated above, then gets supercharged with AI-native design philosophy: instrumented with telemetry, wrapped in evals and tests, provider-neutral, and able to improve with every member interaction. Around it, we build a member experience that matches the caliber of the medicine.
How we would work together
- Engine first. Port the existing knowledge, logic, and workflows into a provider-neutral AI layer, wrapped in tests and telemetry, so the engine becomes stronger, faster to iterate on, and demonstrably correct.
- Founders keep the wheel. The founders remain the product and clinical authority. Weekly working demos of running software keep direction aligned and decisions fast.
- Existing work is the specification. Eight months of product thinking, clinical framing, and lessons learned shape what gets built. The current team’s context is an input we actively want.
Roadmap: public beta in the first week of August
Engine migration and audit
Port coaching logic and workflows; stand up test harness, evals, and telemetry baselines.
Engine running in the new architecture with measurable behaviorConnectors, clinical portal, and experience
Production Junction integration (devices, backfill, lab ordering), portal workflows, refined UI, Expert/Casual modes, security review; closed beta with internal users.
Customer-ready system validated end to end by a closed cohortPublic beta launch
First customer onboarding, monitoring, feedback loop.
Public beta live in the first week of AugustPost-launch hardening
Rapid iteration on live member data, performance tuning, deeper experiments and engagement.
Stability, speed, and polish improving with every releaseSeptember through December
After launch, the focus shifts to fine-tuning driven by real member data: deepening the experiments engine (including CGM), hardening operations and compliance posture with HIPAA-aligned practices, and preparing for scale. Throughout this period, Zhongmin leads interviews to identify and onboard permanent full-time engineers, so that by year-end GetBlue owns both a stable platform and a durable team to carry it forward.
Team, Budget, and Deliverables
The team
Zhongmin Zhu, Ph.D.
Deep expertise in agentic coding and AI-native, production-ready product design; author of the engineering standard described above; built the 48-hour GetBlue demonstration.
Yuri Liang
Led efforts building AI-native tools and support systems for industrial clients including Snap-on and TVEyes.
Elice Li
Five years of backend engineering at major internet companies, plus full-stack app development contract experience.
Steven Sun
Former Apple and Google software engineer specializing in cloud infrastructure.
Operating budget, July–December 2026 (6 months)
| ITEM | BASIS | 6-MONTH COST |
|---|---|---|
| Full-time engineering (2) | $160,000 TC + $20,000 benefits each (annualized), for 6 months | $180,000 |
| Part-time cloud infrastructure | Steven Sun; $80,000 TC + $20,000 benefits (annualized), for 6 months | $50,000 |
| AI model usage (development) | $10,000–$25,000 first month; $5,000–$15,000/month thereafter | ~$50,000 |
| Third-party platforms & tooling | AWS, Junction/Terra, payments, office & collaboration tools, version control, CI | $10,000–$20,000 |
| Total | $290,000–$300,000 |
PM / Director time (Zhongmin Zhu) is contributed to the engagement and not billed within this estimate.
Deliverables
- First week of August 2026: public beta live and first customers onboarding, on a fully functional system spanning the member app, connectors, AI coach, experiments, and clinical portal.
- August–December 2026: continuous maintenance, fine-tuning from real member data, feature depth (CGM, lab ordering), and compliance groundwork.
- By year-end: a stable production platform and permanent full-time engineers interviewed, hired, and onboarded for long-term operation.
This proposal is offered as a partnership. The vision, the clinical depth, and the engine belong to GetBlue. This team brings the AI-native engineering discipline and velocity to carry them to customers on schedule. I would be glad to walk through the live demonstration and tailor any part of this plan. Thank you for the opportunity.
Zhongmin Zhu, Ph.D. / J@polarXmedical.com