Junaid Iqbal | Agentic AI Engineer

We’ve all heard the horror stories: a single decimal point in the wrong place or a duplicate payment that goes unnoticed for months, costing a company thousands. In the world of high-volume transactions, finding these “needles in the haystack” isn’t just a challenge—it’s a necessity for survival.

Whether you’re looking at a retail giant like Daraz or an industrial titan like Aramco, the data doesn’t stop. That’s why I decided to build the Revenue Guard Agent.

The Problem: When Data Outgrows the Auditor

Manual auditing is slow. Worse, as datasets grow, they become “heavy.” If you’ve ever tried to open a massive spreadsheet only to have your laptop freeze, you know the struggle. My goal was to create a tool that was:

1. Lightweight: It had to run on a standard professional laptop (like my i5 Dell) without crashing.
2. Secure: It needed to handle data without being vulnerable to common hacks like SQL injection.
3. Smart: It needed to tell the difference between a “busy day” and a “potential fraud.”

My Method: Simplicity Meets Strategy

Instead of just throwing the data into a basic script, I took a more surgical approach:

• Shrinking the Footprint: I used a technique called Data Downcasting. By telling the computer exactly how much space each number needed, I cut the memory usage by 50%. This means the agent can scan 50,000+ transactions without breaking a sweat.
• Safety First: I integrated a SQLite database using Parameterized Queries. In plain English, this creates a “shield” around the data so that malicious code can’t sneak in and corrupt the system.
• The Math of “Strange”: To find the anomalies, I used the Z-Score (3sigma) method. It sounds fancy, but it basically means the agent looks at what “normal” looks like and only rings the alarm if a transaction is significantly far away from the average.

What’s Inside the Project?

I’ve documented the entire journey. If you’re a hiring manager or a fellow developer, you can dive into the specifics here:

• [PDF Presentation]: A high-level slide deck explaining the business value and technical innovations.
• [ReadMe File]: A step-by-step guide on how the system works and how to set it up yourself.
• [GitHub Repository]: The full source code is available here for review and collaboration.

Final Thoughts

Building the Revenue Guard Agent taught me that the best technical solutions aren’t just about complex algorithms—they’re about making data work efficiently and securely within the limits of real-world hardware.

If you’re interested in financial integrity or Python automation, I’d love to hear your thoughts on the code!