What if AI could turn your application landscape into a living interactive map?
By Neha Devadas, MAM Technology
— 6 minute read
Large organisations often manage complex application landscapes that evolve over time. To support greater visibility and understanding of these environments, we developed an AI-powered detection capability that helps discover, classify and map tools developed by Macquarie employees across collaboration environments. By combining AI with human oversight, the solution demonstrates how technology can help organisations better understand its tools while supporting informed governance and decision-making.
Building a clearer view of the application landscape
As our organisation grows, various tools are developed by Macquarie staff to assist with day-to-day tasks. Understanding what applications exist, how they are used and how they connect to people and processes can help support governance, operational resilience and risk management.
Across large enterprises, business applications play an important role in supporting day-to-day operations. Maintaining visibility of application ownership, dependencies and business purpose is an ongoing governance consideration. Strengthening discovery and classification capabilities can help organisations make more informed decisions about risk, resilience and operational support.
We became fascinated by a simple question: How can organisations improve visibility into their application landscape to support more effective governance?
That question became the foundation for our detection capability, Application Radar.
The solution scans collaboration environments to identify and classify tools containing business logic, using AI to:
- describe what the application does;
- classify it against risk-based criteria;
- identify ownership information;
- and create a centralised inventory of discovered applications.
Importantly, every recommendation generated by our detection capability is reviewed by the application owner. We deliberately designed the solution with a human-in-the-loop approach because governance decisions should never be made by AI alone.
Our goal was not to replace human judgement but to accelerate visibility and provide better information to support decision-making.
Connecting the dots through relationships
One of the most interesting challenges we encountered was understanding dependencies.
Finding an application is valuable, but understanding how it connects to people, locations and other assets is even more valuable.
To address this, our detection capability includes a network graph view that visualises relationships between:
- application owners;
- user-developed applications;
- file locations;
- and connected assets.
This helps provide a richer understanding of application dependencies, ownership relationships and operational connections across the broader technology landscape. Rather than presenting a list of applications, the solution helps create a map of how those applications fit within a broader operational landscape.
We also developed a natural language query feature. Users can ask questions in plain English about the application inventory and receive contextual answers, with relevant applications highlighted within the network graph view.
Technical details
Application Radar is a pipeline, built with the smallest set of components that could solve the problem well.
It begins with a file scanner that works through a SharePoint-style folder and gathers metadata about each file it finds. From there, every file passes through a series of AI-powered stages, each with a single, well-defined job:
- a classifier decides whether a file is a genuine tool developed by Macquarie employees, and how confident it is
- a describer produces a short, plain-English summary of what each application actually does
- an input/output extractor identifies the external files, databases and APIs an application reaches out to
- a risk rater assesses the criticality and business impact of each application
For the final risk tier, the AI provides the raw ingredients (criticality and impact), but the tier itself is derived in code using a simple, transparent scoring matrix, which makes it more explainable for the user.
Every stage communicates through a single, agreed JSON contract. The pipeline only ever writes to it, and the dashboard only ever reads from it. This one-way flow kept our components loosely coupled, which mattered when four people were building different parts of the system in parallel.
We expected the relationship graph to be the hard part, but it turned out we did not need a graph database or any specialist library. Our data was already well-structured, so a plain Python function was enough to build the nodes and edges (connecting owners, applications, file locations and shared assets) and write them straight into the same JSON contract. On the front end, a React dashboard built with Vite uses React Flow to draw the interactive network graph.
The whole system was built and running inside a week, which the small, JSON contract-driven design made possible.
A team effort from start to finish
While the technology was exciting, the most rewarding part of the journey was the collaboration.
I had the privilege of working alongside Tim Chen, Candice Lau and Chinthaka Wimalaratna, who each brought different perspectives and expertise to solving this problem. Together, we were able to rapidly test ideas, challenge assumptions and turn a concept into a working prototype.
Importantly, it reinforced something I have seen time and time again: the best solutions emerge when diverse teams rally around a meaningful problem.
Looking ahead
Application Radar demonstrates how AI can be applied to improve visibility, discovery and understanding within complex application environments. More broadly, it highlights the value of using technology to provide better information, helping people make more informed governance and risk management decisions.