Byzantine Fault Detection in Multi-Agent Reinforcement Learning systems.
Open to research collaborations and engineering opportunities.
Hi, I'm Almond. I engineer AI systems and research complex problems.
My work sits at the intersection of machine learning, reinforcement learning, and applied mathematics + engineering. I build architectures that prioritize technical precision and long-term scalability.
Research & Publications
Policy Graphs for Byzantine Fault Detection in Communicating Multi-Agent Reinforcement Learning
We propose BARD-MARL, a framework that augments Bayesian latent graph learning with policy graph-based behavioral verification to detect and quarantine Byzantine agents in cooperative MARL systems. Achieves F1 = 0.60 at 20% Byzantine fraction — three times the random baseline.
Fault-Tolerant ML
ML-driven replication strategies that enhance fault tolerance in large-scale distributed systems. Published on arXiv.
Mobile DB Edge
Evolution of mobile database technologies from local-first to privacy-preserving edge computing with Federated Learning.
MS in Engineering AI @ Carnegie Mellon University
Researching Byzantine fault detection in communicating multi-agent RL systems. Coursework in ML, DL, AI System Design, Cloud Computing, and MLOps.
Software Engineer Intern @ IBM
Improved GeoSpatial Studio performance by 21%. Built a Selective Component Rebuild Algorithm that increased inference pipeline build speed by 77%.
Software Engineer Intern @ IBM
Built Fine-Tuning and Dataset Factory UIs for Watsonx Geospatial. Engineered a diffing algorithm that optimized fine-tuning payloads by 50%.