Building an AI-powered crossword puzzle with Genkit and Gemini

Google’s I/O Crossword demo, created by Flutter, Firebase, and Very Good Ventures, adds an AI-powered hint system that lets players ask up to 10 yes-or-no questions about a crossword answer rather than directly searching for it. The feature uses the Gemini 1.5 Flash API through Firebase Genkit, Google’s AI integration framework for Node.js, with Go support planned, to manage prompts, model calls, structured inputs and outputs, deployment, and debugging. A versioned dotprompt file defines the model, low-temperature settings, schemas, instructions, and prior question context, enabling the system to return only “yes,” “no,” or “notApplicable” responses. Firebase-hosted Genkit flows use Zod schemas for type-safe data handling and Firebase Anonymous Authentication to limit access to authenticated game users, while Genkit’s local developer UI and tracing tools support testing and iteration. The project also illustrates how changing AI providers or models, such as moving from Gemini 1.0 Pro to Flash or potentially to Agent Platform, can require only limited configuration changes, and its source code and playable game are publicly available.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.