DishGen

Accepts free-form natural language descriptions of available ingredients, dietary preferences, and cuisine preferences, then uses an LLM backbone to generate contextually relevant recipes that match those constraints. The system parses ingredient lists and dietary restrictions from unstructured text input rather than requiring structured form selection, enabling users to describe 'I have chicken, garlic, and need something keto' in conversational language and receive tailored recipe suggestions with ingredient quantities and preparation steps.

Implements a constraint-satisfaction layer that filters generated recipes against user-specified dietary restrictions (vegan, vegetarian, keto, paleo, gluten-free, dairy-free, nut-free, etc.) and allergen profiles. The system likely maintains a mapping of common ingredients to allergen categories and dietary classifications, then validates recipe outputs against these constraints before presenting them to users, ensuring generated recipes do not contain prohibited ingredients or violate dietary rules.

Allows users to specify desired cuisine types (Italian, Thai, Mexican, Indian, etc.) and flavor profiles (spicy, savory, sweet, umami-forward) as input constraints, which the LLM uses to generate recipes that match both the ingredient/dietary constraints AND the culinary preferences. The system likely embeds cuisine and flavor characteristics in the prompt context, enabling the LLM to generate culturally appropriate recipes or flavor combinations rather than generic meals.

Generates recipes with explicit ingredient quantities and serving sizes, and likely supports scaling recipes up or down based on desired serving counts. The system maintains proportional relationships between ingredients during scaling, ensuring that recipes remain balanced when adjusted from 2 servings to 6 servings or vice versa. This is typically implemented through LLM-guided calculation or post-processing of generated recipes to adjust quantities while preserving flavor and texture ratios.

Generates detailed, sequential cooking instructions for each recipe, breaking down preparation into discrete steps with estimated timing for each phase (prep, cooking, resting). The system likely uses the LLM to structure instructions in a clear, beginner-friendly format with explicit guidance on techniques, temperature targets, and doneness indicators. Instructions are generated contextually based on the recipe type and user's implied skill level, potentially including warnings about common mistakes or critical steps.

Tracks user interactions with generated recipes (views, saves, ratings, regenerations) to build a preference profile that influences future recipe generation. The system likely stores user dietary restrictions, cuisine preferences, and past recipe feedback in a user account or session, then uses this history to personalize subsequent recipe suggestions. This enables the LLM to generate recipes more aligned with user tastes over time, avoiding repeated suggestions of disliked recipes or cuisines.

Generates multiple recipes in a single request to support meal planning workflows, allowing users to request 'recipes for a week of dinners' or 'lunch ideas for 5 days' with specified dietary constraints and cuisine variety. The system likely maintains recipe diversity constraints to avoid suggesting the same ingredient or cuisine repeatedly, and may optimize for ingredient overlap to reduce shopping list complexity. This is implemented through multi-turn LLM prompting or batch processing that generates multiple recipes while enforcing diversity and ingredient efficiency rules.

Provides alternative ingredient suggestions when a recipe contains ingredients the user cannot access, does not have on hand, or wants to replace for dietary or taste reasons. The system likely uses the LLM to understand ingredient functions (binder, thickener, acid, fat, protein) and suggests substitutes that maintain recipe balance and flavor. This enables users to adapt recipes to their constraints without requiring manual research or trial-and-error ingredient swapping.

Estimates nutritional content (calories, macronutrients, micronutrients) for generated recipes based on ingredient lists and quantities. The system likely integrates with a nutritional database (USDA, MyFitnessPal API, or similar) to look up ingredient nutrition facts, then aggregates them to provide per-serving nutritional breakdowns. This enables users to track macronutrients for specific diets (keto, high-protein) or manage caloric intake without manual calculation.