After launching a live multiplayer game, players submit hundreds of messy, unstructured bug reports via a Google Form, which dumps raw data into a basic spreadsheet. Previously, a lead developer had to manually read through these rows daily to assess severity and assign tasks—a slow process that wasted valuable engineering time and allowed duplicate bugs to pile up.
To solve this, I built a fully automated, AI-driven pipeline using n8n. This system continuously monitors the database, reads every new bug report, instantly ranks it, checks for duplicates, and creates clean, organized task tickets directly in the team's project board. This removes the sorting delay entirely, allowing developers to focus strictly on shipping fixes.
The workflow initiates the moment a player submits a response through the Bug Report Form. Every submission adds a new row to a Google Sheet. The n8n system constantly monitors this document, catches the new row instantly, and extracts the raw description.
Data Intake Reference:
The raw description is handed over to the Gemini LLM. The AI acts as an automated sorting assistant whose sole job is to interpret the messy complaint and extract three standardized data points:
- Summary: A short, one-sentence summary of the problem.
- Category: The broken game domain (e.g., Gameplay, Art, Network).
- Urgency: The priority level (Low, Medium, High).
The system then fetches historical Trello board data to provide the AI with context on currently known issues.
The AI packages its findings into a strict JSON format, stripping away player emotion or rambling. It then compares the new bug against the historical board data.
- If a New Bug: The system automatically creates a neatly categorized Trello card and fires a "New Bug Alert" to the engineering Discord.
- If a Duplicate: The system halts Trello card creation to prevent board clutter and logs a "Duplicate Alert" in a specific Discord channel.
System Outputs:
The automated pipeline is fully operational, successfully handling real-time data intake, strict JSON categorization via the Gemini LLM, and board-wide duplicate checking. During testing, I successfully fixed empty-payload errors caused by raw text outputs by enforcing strict JSON formatting rules within the Trello integration node.
During deployment, I encountered a critical edge case with AI interpretation: Semantic Over-Matching.
The Break: The AI Agent incorrectly grouped two completely unrelated issues (a player's horse freezing vs. an inventory menu freezing) as a DUPLICATE. The logic failed because the LLM prioritized the players' similar emotional keywords ("freezes," "every time") over the actual broken game mechanics.
The Danger: This is a silent logic failure. The APIs succeed, n8n nodes run perfectly (green), and data successfully routes to the Discord Duplicate log. However, a unique, game-breaking bug is completely hidden from the developers' Trello board.
The Fix: I engineered a strict isolation rule into the AI's system prompt logic. The updated prompt explicitly forces the AI to evaluate the underlying game mechanic rather than matching player sentiment or keyword similarity, ensuring highly accurate duplicate sorting.
With additional development time, the next phase is to build a "Reverse Pipeline" utilizing Trello webhooks. This expansion would automatically notify the original reporting player via Discord or email the exact moment their specific bug card is moved to the "Done" list by a developer, fully closing the community feedback loop.
- Orchestration: n8n (Node-based event-driven automation)
- AI/LLM: Google Gemini API (Prompt Engineering, JSON structuring)
- Integrations: Trello API, Discord Webhooks, Google Sheets API
- Data Format: JSON
To deploy this automation in your own environment:
- Ensure you have a running instance of n8n.
- Download the
bug-pipeline-workflow.jsonfile from this repository. - In your n8n workspace, click Import from File and select the JSON.
- Update the node credentials for your specific Google, Trello, Gemini, and Discord accounts.
- Activate the workflow!