We building an intelligent, cloud backed in-game experience, the idea is simple: bring logic, AI, and data directly into the game, where players and admins interact with it in real time.

We’re implementing a custom minigame experience inside the game, with its own logic layer. This includes:

– Points and in-game currency

– Rewards and progression

– Player state and session tracking

Instead of hardcoding everything, the game logic is designed to be data driven, allowing us to evolve rules and mechanics without redeploying the game itself.

One of the key features is a game admin role. An admin can make live changes to an ongoing game session, such as:

• Switching between day and night

• Assigning weapons, items, or points to specific players

• Influencing the session dynamically without restarting it

Communication between players, the game, and AI agents is handled using a Bot Framework skill.

We’re using Dataverse as our primary database for:

• Game sessions
• Player profiles
• Points, currency, and rewards
• Admin actions and game state

All in-game telemetry is streamed to a data lake. This includes events such as:

• Player deaths and kills
• Building actions
• Monster spawns and movement
• Environmental changes

The goal is to capture everything that happens in the game, at scale, without impacting gameplay performance.

We’re using it in three main scenario where AI is going to play a role:

Game Helper Bot:

Players can interact with a game helper bot through chat. Examples:

• “Build a fortress”
• “Build a wall around me”
• “Create a safe zone”

The agent translates these requests into Minecraft commands that can be executed directly in the game, taking into account:

• The player’s available points or currency
• Game rules and constraints
• Current game state

Player Prompt Improver

Players don’t always write perfect prompts that is why we use an LLM as a prompt improver, which:

• Takes short or vague player input
• Expands and adapts it to the game context
• Produces a richer, more precise prompt for the helper bot

Data Insights Agent

Finally, we’re building a data insights agent on top of the telemetry stored in the data lake.

This agent can answer questions like:

• “What caused most player deaths in this session?”
• “Which areas had the highest monster activity?”
• “How did player behavior change over time?”

What We’re Not Showing (Yet)

We also have plans around:

• Governance
• ALM
• Data indexing and optimization