From Scattered to Structured: A Vision for Automating Architectural Knowledge Management
This paper proposes an automated pipeline to transform scattered architectural knowledge from heterogeneous software artifacts into a structured, consistent knowledge base that supports critical maintenance tasks and natural language querying.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are the head architect of a massive, ever-changing city. This city is your software system. To keep it running, you have blueprints (design diagrams), a list of rules (requirements), the actual bricks and mortar being laid (the code), and a diary of decisions made by the construction crew (documentation).
The Problem: A City of Lost Blueprints
Right now, the information about this city is scattered everywhere. Some rules are written on sticky notes, some blueprints are in a drawer, and the actual construction is happening in a different building.
- The Mess: Because the information is so spread out, it's hard for the workers (developers) to know what the city is supposed to look like.
- The Drift: Over time, the workers build something slightly different than the blueprints say. The diary says "build a bridge," but they built a tunnel. The blueprints aren't updated. This is called "architectural erosion." The city slowly stops matching its original plan, making it hard to fix or expand.
The Vision: An Intelligent City Librarian
The authors, Jan Keim and Angelika Kaplan, propose building a super-smart, automated City Librarian (an AI system) to fix this mess. Their goal is to take all the scattered, messy information and turn it into one clean, organized, and searchable library.
Here is how their "Librarian" would work, step-by-step:
1. The Scavenger Hunt (Extraction)
The Librarian doesn't just read books; it reads everything. It has special tools to look at:
- Text: It reads requirements and meeting notes.
- Code: It looks at the actual construction materials to see what's really being built.
- Images: It even looks at hand-drawn sketches and diagrams.
- Audio: It can listen to meeting recordings to catch decisions made verbally.
It pulls out the important facts from all these different sources, like a detective gathering clues.
2. The Truth Detective (Consistency Checking)
Once the Librarian gathers the clues, it plays detective. It compares the different sources to see if they tell the same story.
- Example: If the blueprint says "Red Door" but the code says "Blue Door," the Librarian spots the conflict.
- It doesn't just find the problem; it tries to figure out which one is right. If it's a small mistake, it might fix it automatically. If it's a big, confusing conflict, it asks a human architect, "Hey, did we change the door color, or is the blueprint wrong?"
3. The Master Map (The Knowledge Base)
All the verified facts are stored in a Master Map (a structured Knowledge Base). This isn't just a pile of papers; it's a connected web where every brick, rule, and decision is linked to the others.
- Because everything is linked, if you want to change one part of the city, the Librarian can instantly tell you which other parts will be affected (Change Impact Analysis).
- It can also check if the current construction follows the original rules (Conformance Checking).
4. The Conversational Guide (Question-Answering)
The best part? You don't need to be a librarian to use this map. You can just talk to the system.
- You ask: "Why did we choose this specific design for the bridge?"
- The Librarian answers: "According to the meeting notes from last Tuesday and the design document, we chose it because it was cheaper and stronger."
- It uses advanced AI (Large Language Models) to understand your questions and find the exact answer in the Master Map, just like asking a knowledgeable tour guide.
5. The Watchful Guardian (Continuous Monitoring)
This isn't a one-time job. The Librarian is an Agent that watches the city 24/7.
- As soon as a worker changes a blueprint or writes new code, the Librarian notices immediately.
- It updates the Master Map in real-time.
- If it gets confused, it pauses and asks a human for help, ensuring that the AI doesn't make a big mistake on its own.
The Challenges (Why it's not easy yet)
The authors admit this is a tough job:
- Too Much Data: The city is huge. The Librarian might get overwhelmed by millions of pages of text.
- Hallucinations: Sometimes the AI might "make up" facts that sound true but aren't. They need to be careful not to let false information into the Master Map.
- Old vs. New: Software changes over years. The Librarian needs to know when to keep old history (for context) and when to throw away outdated info so it doesn't get confused.
- Privacy: If the Librarian listens to private meetings, it must be very careful not to leak secrets.
In Summary
The paper is a blueprint for an automated system that stops software teams from drowning in scattered information. It aims to gather all the messy clues, check them for truth, organize them into a single source of truth, and let anyone ask questions in plain English to understand how their software works. It's about turning a chaotic construction site into a well-organized, understandable city.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.