Muhammad Anas Raza
Ph.D. Candidate & ML Systems / Research Engineer Β· Oakland University
Rochester, MI
Oakland University
I am a Ph.D. Candidate in Computer Science & Informatics at Oakland University, specializing in Machine Learning Systems, Agentic AI, and Deep Learning Reliability.
My work bridges foundational deep learning with high-performance systems engineering, spanning zero-dependency C++/ggml inference engines, multi-backend ML libraries, autonomous multi-agent code generation, and multimodal computer vision forensics.
I am an active open-source contributor to core Keras 3 (20+ merged PRs) and author of timm.cpp (C++ inference engine for 106 vision models), K3Node (multi-backend GNNs with 100% PyG parity), and Keras-Climate.
Featured Systems & Research
View All 12 Projects βtimm.cpp
C++17 / ggml Standalone inference engine executing 106 PyTorch/timm architectures via GGUF with zero Python runtime dependency. Achieved up to 25.8Γ faster cold-start load times via zero-copy POSIX mmap and 3.4Γ to 5.8Γ calibrated quantization weight compression.
K3Node & Keras 3
Core ContributorCore contributor to Keras 3 with 20+ merged PRs. Author of K3Node (multi-backend GNN library with 100% PyG parity across PyTorch, TensorFlow, and JAX with 700+ tests/backend), K3IM (50+ vision backbones), and Keras-Climate.
AutoMAS & NeuroGenesis
IEEE QRS 2026Architected AutoMAS, an autonomous framework translating natural language code intent into working multi-agent systems across 5 cooperative layers (82.75% Pass@1). Authored NeuroGenesis, introducing reasoning-guided multimodal architecture search using LLMs.
DL Reliability & Forensics
Best Poster AwardAuthored RuleBoost (Best Poster Award at IEEE IJCB 2024; 96.19% AUC), LocatorGraph (Empirical Software Engineering 2025; GNN fault localization on DL computation graphs), and DeepTrAnS (IEEE ICSC 2025; 99.88% precision telemetry anomaly detection).
Technical Expertise & Skills
Recent Milestones & News
View All Publications β- 2026: Two research papers accepted at IEEE QRS 2026: AutoMAS (End-to-End Generation of Functional Multi-Agent Systems) and NeuroGenesis (LLM-Driven Neural Architecture Search).
- 2026: Released Keras-Climate, a multi-backend framework for Earth Observation and Climate Modeling.
- 2025: Published Graph Neural Network for Fault Localization in Sequence-Based Models (LocatorGraph) in Empirical Software Engineering (EMSE).
- 2025: Published DeepTrAnS: Mixer-based Unsupervised Time Series Anomaly Detection for Servers in IEEE ICSC 2025.
- 2024: Received the Best Poster Award at IEEE IJCB 2024 for RuleBoost: A Neuro-Symbolic Framework for Robust Deepfake Detection.
selected publications
- AutoMAS: End-to-End Generation of Functional Multi-Agent SystemsIn IEEE International Conference on Software Quality, Reliability and Security (QRS), 2026
- Graph neural network for fault localization in sequence-based modelsEmpirical Software Engineering, 2025
- Multimodaltrace: Deepfake detection using audiovisual representation learningIn Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023