Plotly Technologies Inc.

But in that process, you need to come back with, hey, I learned this lesson.

Hey, I built a Fabric app and it was totally using this D3.js stuff, but I was actually asking it to use Plotly and it broke here.

And I burned a lot of tokens doing this thing and it didn't really work for me.

i think a lot of this stuff is so new that we need a little bit of experimentation time with some experts of your organization or people that can handle this and have them come back with findings and reports and practices that work well for your organization and from those lessons then you start disseminating that knowledge into the rest of the team like hey when you're prompting about a fabric app We need a process to check it.

Like, how do I make this button consistent on all pages? To your point earlier, Kirk, it's wide open.

You wanna use D3.js, you wanna use Plotly, you wanna use-

No.

Um-

And you can specify what kind of library you want.

Do you want Plotly or do you want something else? How do you want it to build visuals for you inside that notebook.

You're right.

It's a lot more verbose on how to, and again, I'm not going to know all that.

That's a, a library that's available there that can then be rendered statically.

You can have interactive canvas or Three.js or Plotly widgets without a Marimo kernel.

So it can be, again, on a static page.

Static reactivity for discrete sliders.

Instead of asking, does this event violate a rule? The AI model asks, does this behavior look abnormal compared with historical patterns? Project architecture the platform was designed as an end-to-end fraud analytics solution.

Technology Stack Component Technology Programming Language Python Machine Learning XGBoost Data Processing Pandas NumPy Model Evaluation Scikit Learn Dashboard Streamlit Analytics Plotly Database SQLite API Layer Python REST Services Data Processing and Feature Engineering machine learning performance depends heavily on feature engineering telecom fraud is rarely identified from a single event the model analyzes behavioral relationships example feature categories traffic behavior sms volume message frequency traffic spikes destination distribution customer behavior account activity usage history transaction behavior network behavior routing patterns time based anomalies sender behavior these features transform raw telecom events into machine learning intelligence Why XGBoost was selected? The platform uses XGBoost as the primary fraud classification algorithm.

XGBoost was selected because telecom fraud datasets typically contain nonlinear relationships, complex behavioral patterns, noisy data, evolving attack methods.

The model works by sequentially improving predictions The simulation output confirms that XGBoost vastly outperformed Random Forest on this specific dataset across both accuracy and efficiency metrics.