Culinary AI Agent

Challenge

A startup approached Hidden Core seeking a technology partner capable of building the backend for an intelligent culinary assistant and diet advisor.
Their in-house team was responsible for developing the iOS and Android app, while our task was to design and implement the AI backend — the technological core that powers the assistant’s reasoning and personalization.

The client’s key goal was to launch the product quickly, while at the same time laying the foundation for long-term independence from third-party AI vendors such as OpenAI, Anthropic, or Google.

Solution

After several technical sessions, we proposed a two-phase implementation strategy:

• For a fast production launch, we selected the Gemini model as the initial LLM and performed fine-tuning to adapt it for culinary-specific tasks.

• We built a vector-based knowledge database that stores recipes, dietary guidelines, cooking techniques, and nutritional information.

• The system’s architecture was designed to easily replace the underlying LLM with a custom model in the future, ensuring vendor independence.

• The next phase involves developing a proprietary AI agent hosted entirely on the client’s infrastructure.

The assistant allows users to:

• Generate personalized meal plans based on preferences, allergies, and dietary needs;

• Receive grocery shopping lists generated automatically from their menu;

• Follow step-by-step cooking instructions with detailed timing and preparation guidance.

Impact

With Hidden Core’s solution, the startup was able to:

• Launch their product rapidly while maintaining a scalable technical foundation;

• Retain full control over their proprietary knowledge base and data pipeline;

• Plan the transition to an in-house AI system without losing momentum or flexibility.

This project helped the client achieve a fast market entry, maintain strategic technological independence, and establish a clear roadmap toward full control of their product’s AI stack.