AI Information & Transparency

Artificial intelligence at Axxa Media is engineered around one core principle: powerful technology should be accessible, practical, and designed around the people who use it.

TasteScribe AI employs artificial intelligence not to replace the human soul of cooking, but to remove kitchen friction—digitizing cherished family recipes, making nutrition information understandable, and organizing groceries. We publish this AI Transparency document to explain how our models work, what inputs are processed, where mistakes can happen, and how user privacy is safeguarded.

Artificial intelligence is applied in specific, well-defined components of TasteScribe AI:

1. Optical Character Recognition (OCR) & Parsing: When you photograph a handwritten recipe card, faded cookbook, or culinary manuscript, our multimodal vision models transcribe visual characters into machine-readable text and separate the output into discrete fields (Recipe Title, Prep Time, Cook Time, Yield, Categorized Ingredients, Numbered Instructions).

1. Generative Recipe Assistance: When you use "AI Recipe Creation," natural language models generate recipe ideas and structured cooking procedures based on your verbal or text prompts (e.g., "Create a Mediterranean lemon chicken dinner using ingredients in my pantry").

1. Healthier Alternatives & Substitutions: When exploring recipe modifications (Low Carb, Low Sodium, Keto, High Protein), the AI suggests culinary substitutions (such as cauliflower mash for mashed potatoes, or nutritional yeast for parmesan) along with ratio adjustments.

1. Nutrition Intelligence Modeling: Our backend computational pipeline calculates approximate macronutrients (protein, carbohydrates, dietary fat) and calories by matching transcribed ingredient lines against nutritional reference tables.

1. "Ask TasteScribe AI" Culinary Chatbot: An interactive conversational assistant that answers practical kitchen questions (e.g., "How do I fix a sauce that broke?", "What can I substitute for buttermilk?").

To deliver these features, TasteScribe AI processes the following inputs:

• User-uploaded photographs of recipe cards, printed books, and plated dishes.
• User text queries, chat questions, and voice transcription inputs.
• Structured ingredient lists and user-selected dietary tags.

Architecture & External Processing Infrastructure: AI requests are routed strictly through authenticated, secure Axxa Media backend servers to enterprise artificial intelligence providers and external processing infrastructure.

• Zero Exposure of Credentials: API credentials and model tokens never reside in client-side mobile code.
• Multi-Provider Capability: Axxa Media utilizes third-party AI technology services, multimodal recognition systems, and generative machine learning models to power OCR and culinary reasoning. Axxa Media does not operate exclusively with any single artificial intelligence provider; underlying models, technologies, and external processing infrastructure may change, be upgraded, or transition over time to ensure state-of-the-art accuracy, resilience, and security.
• Zero Foundation Model Training: Under commercial enterprise terms with our third-party artificial intelligence service providers, user prompts, culinary queries, and uploaded images are processed statelessly and are NOT used to train, fine-tune, or improve public foundation machine-learning models. Your recipes remain your own.

Machine learning is fundamentally probabilistic. Users must be aware of inherent technical limitations:

• Cursive & Handwritten Ambiguity: Cursive scripts, faded ink, vintage shorthand, or stained index cards can result in misread measurements (e.g., mistaking "1 tbsp" for "1 tsp", or "1/2 cup" for "12 cups").
• Regional & Dialect Variations: Specialized cultural ingredients or vernacular culinary terms may occasionally be misinterpreted by the model.
• Altitude & Appliance Variances: AI-suggested baking times and temperatures do not account for your specific oven calibration, altitude, or pan conductivity.
• Hallucinations: While rare in structured recipe extraction, generative language models can occasionally invent cooking steps or hallucinate uncommon flavor combinations.

TasteScribe AI is an informational tool. It cannot smell, taste, or inspect the biological state of ingredients in your kitchen.

You remain solely responsible for:

1. Reviewing Transcriptions: Always cross-reference the digitized recipe against the original photo before cooking.
2. Food Allergens: Never assume the AI detected all potential cross-contaminants or allergen derivatives.
3. Food Temperatures: Ensure all meats, poultry, and seafood are cooked to USDA-approved safe internal temperatures with a calibrated food thermometer.
4. Chemical & Ingredient Safety: Never prepare recipes containing unfamiliar wild foraged mushrooms, toxic plants, or non-edible substances.

We actively monitor algorithmic quality and welcome user reports of inaccurate OCR transcriptions, unsafe recipe instructions, or inappropriate chatbot responses.

To report an AI anomaly or safety concern:

• In-App Feedback: Use the "Report Recipe Issue" link within any recipe detail screen.
• Email Safety Team: Contact [email protected] with the recipe ID and screenshot.
• Account Security Inquiries: For account credential issues or unauthorized access reports, contact [email protected].

We regularly refine our prompt safety guardrails and OCR post-processing heuristics to advance the safety, fidelity, and reliability of our culinary AI.