Project Overview
FoogleTech designed and built a full-stack forensic signature and handwriting analysis platform — a modular, web-based system for document examiners and forensic analysts. The platform combines static handwriting analysis, dynamic kinematic feature extraction, YOLOS-based signature detection, and SigNet CNN comparison — all within a drag-and-drop workflow builder inspired by Flowgram. Built on React.js + FastAPI with Docker deployment on AWS or on-premise.
The Engineering Challenge
- Forensic signature analysis requires multiple distinct AI models — signature detection, comparison, static handwriting analysis, and dynamic feature extraction — each with different frameworks (PyTorch, TensorFlow, scikit-learn) that needed to work together in a unified pipeline.
- The workflow interface needed to be highly flexible — analysts required a visual, drag-and-drop builder (Flowgram-style) to construct analysis pipelines without writing code, a complex UX challenge on top of heavy ML processing.
- Dynamic handwriting analysis (kinematic, spatial, temporal features from stylus/tablet input) required real-time visualization including speed/time graphs, stroke visualizations, and heatmaps — with no existing open-source tooling to build from.
- GPU-dependent models (YOLOS, SigNet, Nithikorn comparison) had to be hosted and orchestrated alongside CPU-only models (handwriterRF, feature extraction), requiring careful DevOps design to manage latency and cost.
- Case management, user authentication (AWS Cognito), report generation (PDF with annotations, CSV exports), and multi-analyst assignment workflows all needed to be built from scratch alongside the AI pipeline.
Our Solution
- Built a Flowgram-inspired drag-and-drop workflow builder using React.js and React Flow — allowing forensic analysts to visually connect analysis nodes (upload → detect → compare → report) without any coding.
- Integrated 6 AI models into a unified FastAPI backend: handwriterRF (scikit-learn, CPU) for static analysis, handwriting-features (NumPy/SciPy) for kinematic extraction, YOLOS-base-signature-detection (PyTorch, GPU) for locating signatures in scanned documents, SigNet (TensorFlow/Keras, GPU) for CNN Siamese signature verification, Nithikorn comparison model (HuggingFace, PyTorch) for two-signature comparison, and HAT for preprocessing and classification.
- Built a scoring engine that produces confidence percentages, match levels, and anomaly flags — giving forensic analysts a clear, defensible output from each comparison rather than raw model outputs.
- Developed real-time visualization using Chart.js and D3.js: speed/time graphs, stroke-level heatmaps, and spatial trajectory visualisations for dynamic handwriting analysis.
- Implemented case management and assignment dashboard, PDF report generation with annotations and graphs, CSV statistical exports, AWS Cognito authentication, and full Docker containerisation for deployment on AWS or on-premise.
Results & Outcomes
- Full-stack forensic analysis platform delivered across 3 phases in 18–20 weeks
- Drag-and-drop Flowgram-style workflow builder for visual pipeline construction
- 6 AI models integrated: YOLOS detection, SigNet verification, handwriterRF, kinematic extraction, Nithikorn comparison, HAT preprocessing
- Confidence scoring engine with match level, anomaly detection, and explainable output
- Real-time visualisation: speed/time graphs, stroke heatmaps, spatial trajectories
- PDF & CSV report export with annotations, case management, and AWS Cognito auth
- Containerised deployment on AWS or on-premise via Docker
Why Signature & Handwriting Analysis System Chose FoogleTech Software
FoogleTech Software is a specialist engineering company with over a decade of expertise in AI, embedded systems, IoT, and full-stack software development — serving product teams and enterprises across AI & Machine Learning and beyond. Our engineers don't just write code — they understand the domain, the constraints, and the real-world pressures that ship deadlines create. For Signature & Handwriting Analysis System, that meant deploying a pre-vetted team with direct experience in React.js, FastAPI, YOLOS, reducing ramp-up time from months to days and delivering production-quality work from the first sprint.
Every FoogleTech engagement starts with a structured discovery phase, follows a disciplined agile delivery model with daily engineering syncs, and ends with complete documentation handover — so your in-house team owns the outcome. No black boxes, no lock-in.