Back

ACUITY · AI Startup Design Hackathon · 2026

Clarity before crisis — turning biometric drift into the next right action

HERO VISUAL — recommended: bracelet motion still on wrist (clean, 16:9)
Use your strongest hardware visual here (bracelet on wrist, minimal background). Alternate option: AI startup title slide.JPG for hackathon context.
Role Systems + product designer (interaction, service, motion)
Team 3 (2 designers, 1 pre‑med)
Timeline 24 hours
AI stack Claude (reasoning), Figma Make (execution), Midjourney (stills), Sora (motion)

Maya wakes up with chest discomfort. Her smartwatch shows an elevated heart rate, but the numbers are still technically within a “normal” range. She hesitates — is this something serious, or just stress?

Modern wearables detect physiological signals but rarely translate them into clear medical action. ACUITY bridges that gap. It detects sustained deviation from a user’s baseline and escalates guidance with proportionate urgency: nudge → schedule → urgent care.

Product

AI-powered preventive health monitoring ecosystem (bracelet + mobile app + provider pathway)

Primary user

High-risk cardiovascular patients during recovery and monitoring periods

Core innovation

An intervention engine translating biometric drift into clear medical next actions

Business model

Distributed through insurers and healthcare systems for preventive monitoring

What I’m optimizing for

Hiring signals for innovation roles: judgment under time pressure, systems thinking, AI orchestration, and end‑to‑end craft. This case study is organized around the decisions that shaped ACUITY — not a checklist of process steps.

The Data‑Action Gap

In our research sprint, the pattern was consistent: people experience symptoms, look at their wearable, and still feel stuck. They have numbers — not meaning. ACUITY is designed as an interpretation layer that bridges raw signals to a confident next step, without alarming users or over‑triaging care.

System at a glance

Inputs

Multi‑modal vitals (HRV, SpO2, sleep, ECG, blood pressure where available) + patient‑reported symptoms.

Brain

Deviation detector compares real‑time patterns against a personalized baseline + clinical risk priors.

Outputs

Tiered guidance: subtle haptic cue → “Schedule now” pathway → urgent escalation when warranted.

Deployment

Distributed through insurers / health systems to reach high‑risk members and align incentives.

System architecture
Wearable sensors (HRV · ECG · SpO2 · Sleep)
Personal baseline model learns normal patterns
Deviation detector identifies sustained drift
Risk classification engine evaluates severity
Intervention engine triggers appropriate action
Product responses: nudge → appointment → urgent escalation

Who

High‑risk cardiovascular patients, with specific attention to groups that are frequently misdiagnosed or dismissed.

What we shipped

Hardware concept + end‑to‑end mobile experience + service pathway (provider + insurer integration).

Constraint

24 hours — forced clarity on the few system behaviors that create trust (thresholds, escalation, language).

PROCESS PHOTO STRIP — pick 3: whiteboard.JPG, me presenting.JPG, team holding certificate.PNG
Use a tight 3‑image strip (same tone, minimal clutter). It signals credibility without turning the page into a scrapbook.

A story‑first walkthrough (lead with motion)

This film is the fastest way to understand the end‑to‑end ecosystem: the bracelet, the tiered alert logic, and the “Schedule now” handoff. Put this above the fold so reviewers get the concept before they decide whether to scroll.

If you export a recruiter cut: 10s (Maya moment) → 15s (tiered alert) → 20s (Schedule now + doctor visit) → 10s (insurer deployment + impact).

Core promise

ACUITY turns sustained baseline drift into a clear action. Not more data. A decision you can trust.

What’s novel

Designing the confidence thresholds and escalation behaviors — the system’s “when do we speak?” logic — as a first‑class UX surface.

DIAGRAM — “System Architecture Glance” (input → model → tier gate → interventions)
Redraw a crisp blueprint diagram from the deck: bracelet sensors → baseline model → deviation detector → logic gate → actions (nudge/schedule/urgent).
HARDWARE MOTION GRID — 4 tiles: wrist close‑up, underside sensors, vibration moment, lifestyle shot
Use your most polished wearable motion assets here. Keep backgrounds neutral; emphasize material/geometry consistency.

AI as a force multiplier (directing > executing)

In 24 hours, the bottleneck is not ideation — it’s synthesis and throughput. I treated AI tools as an orchestration layer: Claude for reasoning and decision framing, Midjourney/Sora for believable media, and Figma Make for rapid UI construction. My job was to set constraints, define evaluation criteria, and keep the system coherent.

Brainstorming · Claude

Problem framing, tradeoff exploration, tier logic, and copy that avoids medical over‑claiming.

Execution · Figma Make

Accelerated layout scaffolds and component variants so we could spend time on the system behaviors.

Media · Sora + Midjourney

Story film + hardware visuals that made the concept feel “real enough” to evaluate.

Key takeaway

AI didn’t replace design judgment — it surfaced more options faster. The differentiator was choosing which system behaviors mattered and making them legible.

RESEARCH PROOF (optional) — screenshot the doc tab list + 1 highlighted excerpt
Only include if it’s curated: a single “research receipt” showing breadth (tabs) + depth (one annotated excerpt) is stronger than dumping many pages.
PROCESS PHOTO — me + krit working desk.JPG (signals collaboration + velocity)
Use one candid work‑in‑progress photo as a transition into decisions and system logic.

The decisions that made ACUITY credible

Decision 1 — Interpretation over more data

We chose to design a translation layer (meaning + next step) instead of a richer vitals dashboard.

In early‑symptom moments, “normal range” numbers can still hide dangerous trend shifts. Users needed contextual guidance, not another chart.

Research repeatedly described uncertainty: symptoms without clarity, or metrics that didn’t map to action.

Copy + thresholds became core UI. We prioritized “what this means” and “what to do now.”

Why

Evidence

Design consequence

Decision 2 — Deploy via insurers, not direct‑to‑consumer

We designed ACUITY as an infrastructure product: high‑risk enrollment + provider pathways + aligned incentives.

The people who most need preventive intervention are least likely to self‑purchase a new wearable. Insurer deployment reaches the right users sooner.

Operationally, prevention works when triage and follow‑through are baked into care pathways — not left to the user.

We added a clinician‑ready summary and a “Schedule now” CTA as the primary action in the alert card.

Why

Evidence

Design consequence

WHITEBOARD ARTIFACT — decision capture: must‑haves/should‑haves (whiteboard.JPG or whiteboard + me.jpg)
Annotate the photo in Figma (light labels) to call out where these decisions were made under time pressure.

Deviation detector + logic gate

ACUITY’s intelligence isn’t “AI magic.” It’s a legible system that weighs drift magnitude, duration, and symptom context. The UX is designed around trust: reduce false alarms, explain why the system is escalating, and always offer a safe next step.

Deviation detector

Continuously compares current signals to a personalized baseline; flags sustained abnormal patterns (not one‑off spikes).

Logic gate

Routes the user to the right intervention tier by combining confidence thresholds + risk priors + user confirmation.

Designing for false positives

We treated “when do we interrupt?” as a design problem: ACUITY escalates only after sustained drift, uses cautious language, and makes “schedule evaluation” the default path before urgent action.

DIAGRAM — state machine: Stable → Elevated → Emergent (include conditions + exits)
Create a clean state diagram with thresholds (duration + magnitude) and safety exits (user says “I feel fine”, or clinician review).

Turning biometric signals into interventions

ACUITY’s intelligence lies in the intervention engine — the system that decides when to act. Instead of reacting to individual spikes, the model evaluates multiple signals together:

  • Drift magnitude — how far the signal deviates from baseline
  • Drift duration — whether abnormal readings persist
  • Context signals — symptoms, sleep, recent activity
Tier System interpretation Product response
Tier 1 Mild deviation Passive monitoring + gentle nudge
Tier 2 Sustained abnormal trend Recommend scheduling evaluation
Tier 3 High-confidence risk Urgent care escalation

Designing these thresholds was the core product challenge: intervene too often and users lose trust; intervene too late and critical events are missed.

Core flow examples

SCREEN RECORDING — “Tier Alert → Card → Schedule Now” (cut from fullWalkthrough.mov)

1) Tiered alert that explains the “why”

The bracelet vibration is intentionally subtle. The primary UI work happens in the card: risk tier, plain‑language explanation, and a single primary CTA: Schedule an appointment soon.

SCREEN RECORDING — “Scheduling → Confirmation → Provider handoff”

2) Low‑friction handoff to care

The product is only as good as follow‑through. The scheduling flow reduces steps, pre‑fills context, and confirms what happens next.

SCREEN RECORDING — “Clinician summary / report view”

3) Clinician‑ready summary

A concise, shareable report translates drift into medically relevant context so providers can triage quickly — without the patient needing to interpret charts.

Note: keep screen recordings framed in a consistent device mock (same scale, same background, no heavy shadows). If you have many screens, group by outcome (alert → action), not by navigation.

User outcomes + system value

ACUITY focuses on reducing the delay between symptom onset and medical evaluation. Even small behavioral shifts could significantly reduce emergency admissions.

If deployed to 100,000 high‑risk cardiovascular patients, modeled outcomes could include:

  • 30–40% reduction in time between symptom onset and seeking care
  • 1–2% reduction in emergency cardiovascular admissions
  • $8–12M potential annual cost reduction for insurers

User impact

Shorter time‑to‑action: reduces the gap between symptom and seeking care.

Lower uncertainty: replaces raw metrics with meaning + a next step.

Less alarm fatigue: escalates only after sustained drift; cautious language; transparent rationale.

Business / system value

Fewer avoidable admissions: even small reductions can meaningfully lower emergency costs.

Aligned incentives: insurer deployment encourages early evaluation and closes the follow‑through loop.

Scalable care capacity: reduces unnecessary ER burden by routing “elevated” to clinics first.

DIAGRAM — “Cost shift” (ER episode → evaluation visit) + “MLR” framing (simple, 1 chart)
Keep this minimal: one bar comparison or Sankey‑style flow. The point is system logic, not precise finance.

Next steps (with more time)

Validate tier thresholds with clinicians, prototype “false positive” handling (user confirmation + clinician review), and pilot insurer enrollment workflows. Expand the hardware story with form‑factor iterations and sensor placement constraints.

Prompt logs + orchestration notes

I include prompt artifacts selectively to show how I direct AI systems — not to show “I used AI.” Keep these collapsed by default so the main narrative stays clean.

Claude · decision framing prompts

Include 3–5 prompts that demonstrate systems thinking and constraint setting. Example categories:

  • Tier escalation rubric (inputs, thresholds, language constraints)
  • False positive / false negative tradeoffs and mitigations
  • Deployment model comparison (DTC vs insurer)
[PROMPT]
Define a 3-tier escalation system for a wearable that detects cardiovascular risk.
Constraints: minimize alarm fatigue; avoid medical over-claiming; require sustained deviation; include a safe, actionable next step.
Output: tier names, trigger conditions, on-device cue, in-app card copy, primary CTA.

[OUTPUT EXCERPT]
…
Sora / Midjourney · media direction

Show how you locked variables (actor consistency, bracelet geometry, lighting, shot constraints) for physics accuracy and reuse.

[PROMPT]
Continue the video using the SAME woman from the previous clip.
Bracelet locked on LEFT wrist in every scene.
Cinematic realism, medium-wide framing, soft natural light.
…
Figma Make · component scaffolding

Include a short log of what you generated vs. what you intentionally edited by hand (copy, hierarchy, states, CTA priority).

[LOG]
Generated: baseline dashboard scaffolds, card variants, spacing system.
Refined manually: alert copy, tier semantics, CTA hierarchy, edge-case states.
END CAP PHOTO (optional) — certificates.JPG or group.PNG (pick one, clean crop)
Optional closer: one photo that signals completion and credibility. Keep it subtle.