Building Guardian AI: A New Frontier for Diagnostic Clarity in Complex Pathology
By Megan Lawther 05.06.2026
I am proud to be part of Google Cloud’s "1,000 Builders, 1,000 Stories."
My journey as a builder started not with a business plan, but with a need for truth. As someone who has navigated years of medical complexity, I realized I had the tools to help myself and others. I am currently developing the Minimum Viable Prototype (MVP) for "Guardian AI," a forensic diagnostic decision-support system designed to navigate complex data challenges in clinical pathology.
The Problem: The Diagnostic Paradox
Navigating complex clinical conditions is a high-stakes riddle. One of the most profound challenges in pathology is that cells of specific conditions can mimic one another at a microscopic level. This mimicry makes the margin for error incredibly thin. Traditional diagnostic pathways often struggle to differentiate between these "master of disguise" pathologies, leading to prolonged diagnostic odysseys.
Furthermore, I have learned that the biggest bugs in our healthcare system aren't always technical; sometimes, they are ethical. Inconsistencies in documentation and the difficulty of verifying clinical accuracy often leave patients in the dark. I am building Guardian AI because the ability to differentiate between these pathways—at a microscopic level—is vital for patient outcomes.
The Solution: Guardian AI (A Work in Progress) Guardian AI is an automated forensic pipeline I am currently developing. My goal is to create a model capable of cross-referencing clinical, laboratory, and imaging parameters to assist in the differentiation of overlapping diagnostic scenarios.
The project is currently in the prototyping and model-building phase. I am refining a proprietary architecture that allows for:
Microscopic Differentiation: Training my model to recognize the subtle, often invisible patterns that separate one clinical pathway from another.
Forensic Data Auditing: Analyzing medical records for integrity, helping to identify discrepancies that could compromise diagnostic accuracy.
Patient Empowerment: Creating a system that puts forensic data analysis back into the hands of the person who owns it: the patient.
Building with the Right Stack: Edge AI, MedGemma, and TensorFlow
To bring Guardian AI to life, I am utilizing the full power of the Google Cloud and Edge ecosystem.
At the heart of Guardian AI is Google’s MedGemma. I chose MedGemma because it offers the specialized medical reasoning and comprehension capabilities I need to parse complex clinical literature. Because it is an open-weights model, it provides the independence I need to build a custom, proprietary pipeline within Google Cloud Platform. This allows me to experiment with my own logic layers and diagnostic workflows without being constrained by "black-box" systems.
My workflow balances rapid prototyping with deep technical control.
I’m using Google AI Studio to test my diagnostic logic, while integrating TensorFlow within Google Colab for the "heavy lifting"—the deep learning required for complex pattern recognition. My technical foundation also relies on the Google AI Edge Gallery and Gemma 4E2B (via AI Core) for real-time, local forensic analysis, ensuring the most sensitive diagnostic auditing occurs locally on the hardware for maximum privacy.
The Vision: Empowerment through Engineering
Building Guardian AI is more than just a coding project; it’s a mission to restore agency to patients. By developing a tool that can provide a second opinion—based on rigorous analysis of clinical data—I hope to help patients navigate their own care with more confidence.
I am at the start of this journey, iterating on my model, refining my feature set, and building a system that values medical truth above all else. I look forward to seeing how the "1,000 Builders" community can help push the boundaries of what’s possible when we combine technical passion with a drive for patient-centric innovation.