Foodhak Assistant

Designing a system that turns health data into daily decisions

Designing a system that turns health data into daily decisions

Shirley's Scope

0 - 1 product, AI-driven health platform

Personalization

Trust

Behavior change

The Problem

Health apps today fail at behavior change

Too much data, not enough guidance

Generic recommendations → low relevance

Users drop off due to fatigue + confusion

Ideally I'd like to make health conscious feel effortless, not overwhelming

Shirs

Design Principle

From tracking → guiding

Instead of:

  • Logging calories

  • Calculating calories and stats

We shift to:

  • Contextual insights

  • Actionable recommendations

  • Behavioral nudges

Core Concept

Foodhak Assistant Insights

“You’ve had high sodium for 3 days — here’s a small change today.”

“You’ve had high sodium for 3 days,

here’s a small change today.”

Interprets user behavior (food, sleep, activity)

Surfaces insights at the right moment

Explains why recommendations are made

System Thinking

Closed-loop design

User data input —> Processing (AI engine, RAG controlled env.) —> Contextual recommendations

Key Product Decisions

Adaptive Experience

Adaptive Experience

Not static plans, reduce cognitive load

Proactive insights

Proactive insights

AI surfaces weekly progress without prompting

Value adherence pattern

Value adherence pattern

Adapt to real behavior, not ideal scenarios

Explainability over “magic”

Explainability over “magic”

Recovery needed

User Impact

  • Improved adherence to routines

  • More stable weight management

  • Reduced burnout from tracking

  • Better recovery and sleep awareness

Platform Impact

  • Scalable personalization without coaching

  • Strong privacy moat (on-device data)

  • Clinically defensible system

  • Foundation for long-term AI health optimization

Business Impact

  • Clear differentiation from calorie trackers

  • Premium personalization tier

  • Pathway to clinical + enterprise partnerships

  • Potential for publishable health outcomes

What I'd Improve Next

In health, precision is valuable

Deeper Personalization Using Longitudinal Data

Deeper Personalization Using Longitudinal Data

Not reacting to today → learning patterns over weeks/months

Examples:

  • Pattern Detection

    “You consistently feel fatigued on low-carb days after 3PM”

  • Adaptive Recommendations

    • Week 1: “Reduce sugar intake”

    • Week 4: “Switch lunch composition — your energy drops after this specific meal pattern”

  • Personal Baselines (not generic)

    • Your “normal” sleep = 6.5h

    • Your “optimal” = 7.2h (based on recovery data)

Comprehensive Visualization

Comprehensive Visualization

Charts ≠ understanding

Allow Predictive Insights

Allow Predictive Insights

“What Happens If…”

Examples:

  • Order Detection

    "You ordered food from Doordash.."

  • Adaptive

    • Quick log?

  • Prediction Baseline

    • User eats lunch at 12PM - 1PM everyday.

    • You consistently snack at …

    • You might feel, want to …

Deeper Personalization Using Longitudinal Data

Not reacting to today → learning patterns over weeks/months

Examples:

  • Pattern Detection

    “You consistently feel fatigued on low-carb days after 3PM”

  • Adaptive Recommendations

    • Week 1: “Reduce sugar intake”

    • Week 4: “Switch lunch composition — your energy drops after this specific meal pattern”

  • Personal Baselines (not generic)

    • Your “normal” sleep = 6.5h

    • Your “optimal” = 7.2h (based on recovery data)

Comprehensive Visualization

Charts ≠ understanding

Allow Predictive Insights

“What Happens If…”

Examples:

  • Order Detection

    "You ordered food from Doordash.."

  • Adaptive

    • Quick log?

  • Prediction Baseline

    • User eats lunch at 12PM - 1PM everyday.

    • You consistently snack at …

    • You might feel, want to …

AI is only valuable when grounded in real user context


Trust is also a design problem, not just a technical one


Constraints (clinical, privacy) can strengthen product decisions

Retrospect, Shirs

shirleythiswillwork.today