Inspiration
World of Warcraft gives players a huge amount of useful data, but turning that data into a clear answer to “what should I do next?” can still take a lot of manual searching.
ARTOMIC WoW Agent was built to reduce that friction.
What it does
ARTOMIC connects to live World of Warcraft character data and turns it into actionable progression guidance.
It can:
- Read the player's live character and equipped gear through the Blizzard Battle.net API.
- Detect the weakest equipped gear slots.
- Search the current dungeon pool for relevant loot candidates.
- Rank general Fire Mage stat-profile options.
- Recommend a specific dungeon to farm.
- Show the bosses and items that justify that recommendation.
- Compare the player's equipped item level against the expected reward level for the selected dungeon difficulty.
- Track achievement progress.
- Provide Dungeon Intel.
- Block or qualify recommendations when data is stale or cannot be sufficiently verified.
Core principle
It does not pretend to know what it cannot verify.
ARTOMIC separates source-verified facts, general heuristics, and areas that would require a personal simulation.
Example
For the current test character, ARTOMIC detected ilvl 174 shoulders as a weak slot.
It scanned the configured dungeon pool, found 13 source-verified candidates and recommended Skyreach, where two A-profile shoulder options are available across two bosses.
For Mythic 0 it then compared:
- Equipped shoulders: ilvl 174
- End-of-run reward: ilvl 246
- Difference: +72 ilvl
Result:
YES — ILVL UPGRADE IF THIS ITEM DROPS
How I built it
The current prototype is a Python desktop application with a Tkinter interface.
It integrates:
- Blizzard Battle.net API for live character and equipment data
- Warcraft Wiki / MediaWiki API for additional source verification
- Gemini 3.5 Flash through the Google GenAI SDK
- A rule-based evidence and season guard layer
- A recommendation layer that converts raw loot data into a practical farming decision
Challenges
One of the hardest parts was avoiding confident recommendations from incomplete data.
Rather than allowing the AI layer to invent certainty, ARTOMIC uses explicit guards for season dates, sources, item-level data and recommendation confidence.
Another challenge was turning raw loot lists into an actual decision. The prototype evolved from simply displaying candidates to recommending a dungeon based on the concentration of strong general stat-profile options.
What I learned
Building ARTOMIC showed me that an agent becomes much more useful when it does not merely return information.
The useful step is converting information into:
Here is what I know → here is why → here is the next action.
What's next
The next steps are:
- Google Cloud deployment
- deeper Gemini agent orchestration
- persistent player preferences and context
- broader specialization support
- live season refresh workflows
- stronger simulation integration
- more autonomous progression planning
ARTOMIC WoW Agent is currently a working local proof of concept.
This is an unofficial fan-built project and is not affiliated with or endorsed by Blizzard Entertainment.
Log in or sign up for Devpost to join the conversation.