Violet Olds

Issue 06

Meet Kyle.

A health companion built around a radical technical premise: your body gets a vote, and your data stays yours.

KYLE09:41

Today's lane

Steady.

Move with intention. Eat something useful. Rest still counts.

SleepIN RANGE
EnergyLISTEN IN
MovementFLEXIBLE
ILLUSTRATIVE DATA ONLY

Observe / 30-day baseline

Signals, not verdicts.

Kyle separates measured, entered, derived, and AI-generated information so a suggestion never masquerades as a fact.

NO SHAME. NO DIAGNOSIS. NO INVISIBLE INFERENCE.

Private prototype / Product case study

Care is a systems requirement.

Kyle is a connected Android and web prototype for understanding everyday health signals without turning a person into a performance dashboard.

It combines consented Health Connect data, deterministic observability, adaptable coaching, kitchen support, and explicit AI features. The voice is supportive gym-bro fluency powered by a matriarchal operating system: care is strength, rest is part of training, and macho optimization theater is cancelled.

Why there is no live demo: Kyle is a private health-data prototype. This case study uses illustrative interface content and intentionally does not expose the dashboard, credentials, or personal wellness information.

The governing rule

The person is the authority over their data.

Privacy is not compliance copy at the bottom of the page. It shapes what Kyle collects, where computation happens, what crosses the network, how AI is introduced, and whether a feature ships at all.

How the system works

Local first. Explicit bridge. Useful cloud.

01

On the device

The Android companion reads only approved Health Connect categories. It calculates coverage, baselines, thresholded shifts, and early correlations from daily aggregates. Core observations work without an AI model.

  • Permission-by-permission access
  • Deterministic signal engine
  • Offline recipe fallback
  • Sensitive categories stay local
02

The consent bridge

Nothing silently drifts into the cloud. A person opens Sync and deliberately sends up to 30 days of daily rollups—not raw samples, medications, notes, or background activity.

  • Manual sync
  • Minimum necessary fields
  • Purpose-bound consent
  • Clear failure states
03

On the web

The private dashboard makes patterns legible. Server-side AI is reserved for features with a concrete benefit: structured recipes, coarse workout adaptation, and explicitly requested weekly reports.

  • Private authenticated surface
  • Secrets stay server-side
  • AI output is labeled
  • Deletion and export matter

Engineering decisions

More restraint. Better product.

Today is incomplete
Current-day data is excluded from anomaly detection, preventing a partially synced day from becoming a false health insight.
AI is a feature, not the foundation
Baselines and shifts are deterministic. Declining AI does not break core tracking, review, export, or deletion.
Sensitivity has tiers
Cycle, fertility, pregnancy, sexual-health, medications, mental-health, location, and clinical records require stricter boundaries than ordinary wellness aggregates.
Fallbacks tell the truth
If a recipe request cannot reach the model, Kyle labels and serves an offline library match instead of pretending the generated path succeeded.

The stack

Native where it matters. Full-stack where it helps.

This is an actively evolving private prototype, not a medical device. It supports reflection and care; it does not diagnose, prescribe, or replace a clinician.

Back to the receipts

One project. Two product surfaces. Zero surveillance business models.

See all selected projects