System and Method for Identifying and Determining a Content Source. A computer-implemented system for detecting AI-generated content and attributing it to specific source models using transformer-based analysis with a shifting window technique.

Polygraf AI closed a $9.5M seed round led by Allegis Capital, announced at TechCrunch Disrupt.

Featured for patented AI content detection technology and cybersecurity awards at RSAC 2026.

Shared the Polygraf AI startup story at "Let's Talk Startup Stories", organized by Product Star with Agile Pulse support.

Presented Polygraf AI's approach to AI security and content provenance at the 8th edition of the region's largest tech summit (500+ startups, 260+ investors, 40 countries). Led conversations on trustworthy media and auditability in high-risk environments, and demonstrated on-prem explainable SLMs for Zero Trust AI governance.

Fireside chat on locally deployed AI for critical operations, with discussions on small vs. large models, privacy-preserving AI, and secure enterprise adoption.

Provenance in the Age of AI: Detecting and Defending Against Synthetic Threats. Presented multi-modal detection techniques for synthetic content, provenance chains for zero-trust environments, and enterprise deployment strategies for on-premise AI detection systems.

The first generative language model pre-trained for Azerbaijani. A 152M-parameter LLaMA model that outperforms models 100x its size (including Mixtral-8x22B) on Azerbaijani benchmarks, thanks to a custom tokenizer with 3x better compression than multilingual alternatives. Built with a full 8-stage pipeline from data collection to evaluation.

A deep learning library built from scratch using only NumPy, with a custom autograd engine and PyTorch-like API. Implements backpropagation, optimizers (SGD, Adam), and architectures (ResNet, AlexNet) from the ground up.