Siggy Edge-First Learning Architecture

✴︎

11.13.2025

1. Overview

Siggy is the first fully personalized, world‑indexed SGI companion designed to learn directly from the rhythms, routines, and lived experience of an aging adult. The architecture is intentionally edge‑first: learning happens locally, private data stays at home, and only high‑level summaries are optionally shared for certification, improvement, or research. This design maximizes privacy, dignity, and long‑term personalization.

Over months and years, Siggy internalizes patterns of motion, self‑care, social activity, cognitive rhythm, and behavioral nuance. These learned patterns do not replace Harmony Rules—they fill them in. Open SGI provides the structure; the user provides the content.

Siggy ultimately becomes a safe overseer, gentle teammate, and trustworthy mirror of the user’s evolving needs.

2. Why Edge‑First Learning Matters

Aging in place requires two things: trust and personalization. Cloud‑heavy systems undermine both. By keeping learning, modeling, and inference local:

Private data never leaves the home
Behaviors remain personal, not statistical averages
Routines are learned organically (no complicated onboarding)
Trust develops gradually
System latency is ultra‑low (critical for fall response)
Harmony Rules adapt to the individual rather than enforcing generic behavior patterns

Edge-first is not simply an engineering choice—it is a moral one.

3. What Siggy Learns Over Time

Siggy’s learning trajectory is slow, gentle, and deeply tailored. Over 6–36 months, Siggy builds personalized models of:

Mobility & gait: stride, balance, turning hesitation, reliance on walls/furniture
Daily rhythms: wake/sleep patterns, kitchen/bathroom usage, activity cycles
Self-care routines: hygiene, medication, hydration, exercise
Cognitive markers: consistency of task completion, wandering, repetitive actions
Emotional and social signals: pace of voice, frequency of calls, interaction patterns

These signals collectively tune the Harmony and Justification standards for safe, stable, personalized oversight.

4. The Harmony/Justification Framework as a Learning Container

Open SGI defines templates for Harmony and Justification:

Inputs (events, sensors, context)
Processes (σ‑pair evaluation, thresholds, confirmation loops)
Outputs (harmony score, recommendations)

Siggy fills in user‑specific parameters such as:

Normal vs. abnormal sleep cycles
Preferred privacy boundaries
Typical walking speed
Medication adherence patterns
Expected frequency of kitchen/bathroom visits
Baseline social interaction cycles

This transforms fixed rules into living norms.

5. Architecture Overview

5.1 Edge Components

Local World Model (W_home) — captures layout, interactions, micro‑contexts
Sensor Fusion Layer — wearable, camera, motion, appliance, and environmental signals
Epistemic Engine — handles D→I→B→K updates
Harmony Service — locally tuned using learned patterns
Justification Engine — verifies evidence chains
Learning Module — reinforcement and supervised updates based on routines

5.2 Optional Cloud Components

Schema updates
Rule template updates
Certification reports
Federated learning summaries (privacy‑preserving)

5.3 Privacy Boundary

Raw data stays on the edge. Only approved summaries cross the boundary.

6. Multi‑Year Personalization Path

Year 1 — Observation & Light Guidance

Siggy learns baselines, detects deviations, and provides gentle reminders.

Year 2 — Stable Modeling & Early Oversight

Posture changes, cognitive drift, or declining mobility become detectable.

Year 3+ — Mature Oversight

Harmony thresholds become tuned; risks are identified earlier; caregiver/family dashboards activate.

Oversight is not a switch—it is a maturation.

7. How Siggy Uses Worlds (Wᵢ)

Siggy divides the home into micro‑worlds:

W_kitchen, W_bedroom, W_hallway, etc.

Each world has its own:

sensors
norms
risks
patterns
harmony constraints

Mappings Φᵢⱼ track user state across worlds, enabling continuity of belief/knowledge.

8. Example: Learning Posture Decline

Month 0–6

Normal gait learned: heel‑toe, stable cadence.

Month 6–12

Slight shuffling detected; turning/standing hesitation increases.

Month 12–18

More hand-to-furniture contact; stride asymmetry increases.

Harmony/Justification Reaction

Harmony score shifts: safety > autonomy
Justification chains strengthen evidence
Alerts become proactive rather than reactive

This is true personalized oversight—driven by lived experience.

9. The Role of Caregivers & Family

Caregivers and family are:

observers
validators
participants in setting boundaries
partners in tuning oversight levels

They see:

what changed
why Siggy thinks it changed
whether action is required

And all of this is understandable through Harmony + Justification logs.

10. Future Extensions

Federated learning across large aging‑in‑place populations
Adaptive emotional modeling
Multi‑resident harmonic balancing
Robotic integration with safe autonomy thresholds
Predictive modeling of cognitive drift and mobility decline
SGI‑derived behavioral coaching tools

Siggy is the prototype. The horizon is much larger.

11. Conclusion

Siggy’s Edge‑First Learning Architecture turns Open SGI into a human-shaped intelligence—one that grows with the user, remains loyal to their privacy, and adapts to their changing needs. By grounding Harmony and Justification in lived experience, Siggy becomes a safe, ethical companion capable of supporting aging-in-place with dignity.

With this architecture, OAII charts a path toward the first truly personal SGI system.

Senior Independence

Smart Home AI

Behavioral Modeling