It started with a small moment—Amy Marks tried to deposit a check through her banking app, paused, and asked: “How does the bank know if this check is real?”
That question sparked the idea behind Document Analysis Toolset, her solo submission to the HP + NVIDIA Hackathon. Over two months, Amy transformed that spark into a real-time AI-powered platform that helps detect document tampering, signature forgeries, and more.
Her mission? Make document forensics fast, accessible, and AI-powered.
Solving a Real Problem
Document fraud costs organizations billions annually. Forged signatures, manipulated PDFs, and subtle edits can go undetected through manual review.
Amy’s project tackles this by automating three forensic capabilities:
-
Signature Detection using YOLO-based object detection
-
Signature Verification through deep learning classifiers with confidence scoring
-
Manipulation Detection with Error Level Analysis to flag digital tampering
The tool works in real time, supports multi-page documents, and presents results in a clean, user-friendly interface—no forensic background required.
“I built this for people who need expert-level results, without needing expert-level training.”
️ Built with HP AI Studio
Amy built the project using:
-
HP AI Studio for model training, iteration, and deployment
-
MLflow to serve models directly to her Streamlit UI
-
NVIDIA RTX 4070 Ti Super GPU on Windows via WSL
-
Custom Python classes for MLflow compatibility
“HP AI Studio felt like Docker for AI—but even more seamless. MLflow made experimenting and deploying models fast and smooth.”
What’s Next?
Amy isn’t done. She’s exploring ways to:
-
Use pen stroke and pressure data to improve signature verification
-
Continue tuning models for real-world accuracy
-
Expand the tool’s use in legal, financial, and compliance contexts
“No system is perfect, but I want to get as close as possible—and make it practical for everyday users.”
A Project with Real Impact
Amy turned a real-world curiosity into a powerful AI solution. Her project proves how a focused idea, smart tooling, and a curious mind can build something that solves real problems—beautifully.
“This was one of the most fun hackathons I’ve done. I learned a ton—and I’m still building.”
Try It Out
-
Live App: doctools.streamlit.app -
Devpost: Document Analysis Toolset -
GitHub: github.com/msamylea/hprepo -
Interview Recording: Watch on Zoom -
Want to try HP AI Studio for yourself?
Start for free