← Back to Home

AI Transparency Statement

At Tiny Tastes Tracker, we believe that parents using intelligent tools deserve full transparency about how artificial intelligence is integrated into the products they trust with their child's data.

Last Updated: July 15, 2026

This statement details how and where we use AI, what models we use, and our commitment to safety.

1. Where We Use AI

Tiny Tastes Tracker integrates artificial intelligence into the following specific features:

• "Sage" Copilot: The interactive chat assistant that helps you plan meals and understand developmental trends.
• Sleep Prediction: Analyzing trailing sleep logs to estimate the next optimal nap window.
• Picky Eater Strategies: Generating customized, step-by-step "Food Bridging" therapies based on foods your child loves and rejects.
• Flavor Pairings: Suggesting creative ingredient combinations.
• Data Log Ingestion: Processing natural-language meal descriptions or photos into structured food metrics.
• Label Reference (Medication): Surfacing general product label information as a logging convenience, not a safety assessment.
• Search-Grounded Answers: Sage answers are grounded via live Google Search to surface real, verifiable sources. Source links are displayed alongside answers so you can verify information directly.

2. The Model We Use

We utilize Google's Gemini 2.5 Flash / 2.5 Pro via the official API to power all AI interactions. We do not use proprietary, black-box models developed completely in-house.

3. How We Protect Your Data

When you engage with an AI feature, only the necessary subsets of your child's data (e.g., age in months, the last 48 hours of sleep logs, or a specific photo) are transmitted to the Google Gemini API.

Our Commitment to Data Privacy: We use the Google Gemini API. Per Google's current API policies, data submitted via the API endpoints we use is not used to train Google's underlying foundation models. However, Google's terms evolve over time. We encourage you to review Google AI's Terms of Service for the most current and authoritative information. We are actively evaluating upgrading to Google Cloud Vertex AI endpoints with an explicit Data Processing Agreement, particularly important given that we transmit children's health data.

4. Human-in-the-Loop & Deterministic Overrides

We recognize the limitations of Large Language Models. To mitigate risks:

• Deterministic Allergen Overrides: Before the AI returns a recipe or recommendation to you, it passes through our local codebase's deterministic (non-AI) safety circuit. If a known severe allergen is detected in the response, the app actively filters or flags the output.
• Strict Prompt Guardrails: Our system prompts are hard-coded to intercept medical questions and force the AI to return a "Consult your pediatrician" disclaimer instead of attempting to diagnose a condition.
• Medication Label Reference (Not Analysis): The medication feature is explicitly forbidden from rendering safety verdicts, dosage comparisons, or clinical determinations. It surfaces publicly available product label information only.
• Live Source Grounding: Sage's answers are grounded with live Google Search results, providing real citations instead of AI-fabricated references. Source chips displayed in the app link directly to the original web pages, as required by Google's grounding terms of service.

5. Built-In Limitations

AI models operate on probability, not medical certainty. They lack common sense and clinical judgment. The outputs presented by our AI features must be treated as ideas and suggestions, not facts. We actively encourage our users to cross-reference AI-generated recipe substitutions or sleep strategies with authoritative sources like the American Academy of Pediatrics (AAP).