Privacy-first ear biometric segmentation - 99%+ accuracy with <2M parameters for edge authentication and GDPR compliance

A stateful AI agent framework powered by the Cognitive Lattice to solve complex tasks with persistent memory and reliable tool orchestration.

A curated collection of privacy-preserving machine learning techniques, tools, and practical evaluations. Focuses on differential privacy, federated learning, secure computation, and synthetic data generation for implementing privacy in ML workflows.

Production Android AI with ExecuTorch 1.0 - Deploy PyTorch models to mobile with NPU acceleration and 50KB footprint

Privacy-first decentralized AI training network combining federated learning, blockchain incentives, and quantum-safe cryptography. Enable secure collaborative model development without sharing raw data.

Federated training on MNIST with differential privacy noise + FL metrics tracking

Build a decentralized AI infrastructure on Solana, enabling secure on-chain model training and creating a global marketplace for AI inference services.

Secure, local-first workbench for refining LLM prompts. Features PII sanitization, Shadow Model architecture, and BIP39 encryption. No data leaves your machine.

Agentic digital health assistant, powered by Federated Learning, autonomously supports patient recovery post-discharge while preserving privacy across clinical institutions.

Privacy-preserving federated learning pipeline built with TensorFlow Federated. Simulates multi-client local training with centralized weight aggregation (Federated Averaging), achieving ~96.7% global accuracy on MNIST. Showcases secure, scalable distributed AI model training without raw data sharing, adaptable to real-world sensitive environments.

A decentralized, diffusion-based U-Net framework for privacy-preserving brain tumor segmentation from MRI images.

A Modular Knowledge Transfer System for Large Language Models

Implementation of Federated Unlearning for medical image classification using the FedEraser approach. Demonstrates how client data contributions can be removed from a trained federated learning model without full retraining.

A federated healthcare triage assistant that routes patients to appropriate care while ensuring privacy and reducing health disparities via multi-stakeholder governance.

AgisFL is a cutting-edge, production-ready cybersecurity platform that combines advanced federated learning with real-time threat detection to provide comprehensive network security monitoring and incident response capabilities.

GPT-OSS B20 Local Execution. Lightweight local environment for running it with Python 3.12 and CUDA acceleration. - Run GPT-OSS B20 entirely offline - Optimize text generation with GPU - Enable fast, secure inference on consumer hardware.

🤝 Enable federated AI and compute sharing while preserving privacy, empowering users to control their data and collaborate on decentralized models.

A privacy-preserving Federated Learning implementation on CIFAR-10 using PyTorch and PySyft. Simulates distributed training across 10 virtual workers with secure model aggregation.

A vision for an open, democratic AI infrastructure — where individuals and communities share knowledge and compute without losing autonomy or privacy.