When a user sends a vague request like “collect 20 stone blocks”, how does a Minecraft bot know what to actually do?
Answer: We let AI figure it out.
We’ve integrated Azure OpenAI into our automation pipeline to break down high-level goals into executable task sequences. The bot doesn’t just blindly follow commands… it plans!
The Planning Pipeline
When a goal arrives via our Logic App:
Azure Function receives the goal
OpenAI breaks it into step-by-step tasks
Service Bus queues the planned order
Minecraft Bot executes using A* pathfinding
User: “collect 20 stone blocks” ↓ AI Planner → tasks: [{ type: “collect”, params: { itemType: “stone”, quantity: 20 }}] ↓ Bot executes with A* navigation algorithm
Our bot uses implements the classic A (A-star) search algorithm* to navigate the Minecraft world:
A* finds the optimal path to destinations
BFS (Breadth-First Search) locates nearby blocks to mine
Stuck detection with automatic recovery maneuvers
Self-preservation system monitors health, oxygen, and hostile mobs
The AI plans the what. The algorithms handle the how.
Instead of hardcoding every possible command, we let the LLM interpret natural language and generate valid task sequences. The bot becomes smarter without more code.
Our solution demonstrates how multiple Microsoft Cloud APIs can be combined into a single, event-driven architecture, with Dynamics 365 Finance & Operations (FO) acting as the production system of record.
The flow starts when a production order is created in D365 FO. FO exposes the order through its OData / Business Events APIs, which are consumed by an Azure Function. This serverless component validates and transforms the data, ensuring the core ERP system remains isolated from downstream processing.
From the Azure Function (dataverse API), the production event is published to Azure Event Hub, using Azure’s native messaging APIs. Event Hub enables reliable, scalable streaming and decouples FO from consumers, allowing the architecture to grow without impacting the ERP system.
The streamed events are then ingested by Microsoft Fabric, where data pipelines process the production data in near real time. Fabric acts as the central data and analytics layer, feeding insights to Power BI dashboards and triggering actions in the Power Platform. In some scenarios, enriched data can also be written back to D365 FO through its APIs, closing the loop.
By combining 1) Dynamics 365 FO APIs, 2) Dataverse API and Event Hub APIs, and 3) Microsoft Fabric and Power BI APIs, this architecture shows how Microsoft cloud services can work together. earning its place on the Stairway to Heaven.
We couldn’t let another team miss out on the awesome solution from one of our sponsors, OneFlow. So when Evidently Crafted needed help, our very own Sebastian stepped in and helped them get access. Anytime!
A low-code Power Apps canvas app paired with a pro-code PCF component. The canvas app manages configuration and UX, while the PCF pulls real-world gold prices from a third-party API and converts NOK into Minecraft gold nuggets, bars, and blocks.
Parameters flow cleanly between low code and pro code giving them a strong relation between the two.
In this solution, we also use external APIs #Dataminer
The API we use comes from an external API which gathers external data from exchange rate to add real life value to the resources we gather with our BOT;
Whoever decided that Minecraft had to be blocky and ugly clearly never tried hard enough.
For our solution, we wanted more than just a functional interface. we wanted something you actually enjoy looking at. ooooor something that just pops in your face. Something shiny. Does it add any function? no, but it looks nice.
Is it a little chic? Absolutely. Is it almost too pretty?
Maybe….Pixels well spent. ✨
Plug N`Play
AAAAND, since this mod is a custom plugin we modified for Minecraft (microsoft owned) we can also claim the Plug N’ Play badge 🙂
Whats the business need? making the product pretty to look at of course!
if no one wants to look at it, no one buys it…and thats the truth
We may not have the coziest camping gear, fancy gadgets, or enchanted sleeping bags… but we do have a disco light. And frankly, that’s all the warmth a group of committed crafters need. 🕺✨
While others retreated to their beds, we stayed…brains buzzing, disco glow pulsing.
We are proud to say we’re the only team with all participants still hacking past 01:00 at night, powered by caffeine, code, and each other’s company.
Happy campers aren’t defined by equipment.
They’re defined by team spirit, stubborn dedication, and enjoying the madness together (even it it`s late at night)
Day one started the only way a proper hackathon should: with hope, coffee, and the unshakable belief that this will definitely work.
Our mission is simple to explain and slightly insane to execute: connect real production orders from Dynamics 365 Finance & Operations to a Minecraft world, where they show up as structured tasks for an AI-controlled NPC robot worker. When the ERP says “produce,” the bot mines, harvests, crafts, and delivers. When the blocks are ready, the order flows back and closes in F&O. ERP in. Blocks out. Loop closed (eventually).
What We’ve Built So Far (Day 1 Reality Check)
We stood up a Java-based Minecraft server, hosted on a Raspberry Pi, because why make things easy when you can make them interesting. A backend service now handles communication between the game world and the outside systems, and this same backend is already feeding data into Power BI for dashboards that tell us how hard the world is working.
The NPC bot is alive (mostly). It can connect to the server, understands the idea of work, and is slowly being taught the difference between “mining resources” and “wandering off into the sunset.”
On the enterprise side, we’re running F&O in a DevBox, experimenting with triggering production orders through contracts and approval flows, and wiring up integrations that will eventually let real business demand drive in-game behavior. DevOps is hosting the server repo, VS Code is hosting our sanity, and we are focusing on implementing Oneflow i in the approval flow, waiting to be fully summoned.
So far this is a real business flow, just expressed in blocks. Production orders, bills of materials, automation, feedback loops There’s no drag-and-drop magic holding this together. Java servers, backend services, bot logic, and integrations are all very much hand-crafted and slightly dangerous. Server and bot telemetry is already flowing into Power BI, turning block-breaking into charts, KPIs, and suspiciously professional dashboards. We’re deliberately mixing pro-code foundations with Power Platform components where it makes sense. A modern ERP logic on one side, rapid configuration on the other.
Even when your worker is a Minecraft robot, questions about autonomy, control, and feedback loops matter. We’re designing this with clear system boundaries and responsibility in mind, and keeping the bot connected only to Minecraft ruleset.
Status at End of Day 1
Is it finished? Absolutely not.
Is the direction clear? Very much so.
Hope still thrives. And the pickaxe is not broken yet. 🪓✨
