Cheng Fang

AI Research Scientist | Machine Learning Engineer

Ph.D. in Computer Engineering, Rutgers University

Robust decentralized ML, RAG systems, agentic AI, and LLM evaluation

fangchengpeter@gmail.com | LinkedIn | GitHub | CV / Resume

About

I am an AI Research Scientist at Knova AI and a Ph.D. graduate in Computer Engineering from Rutgers University. My work focuses on robust decentralized machine learning, statistical learning, RAG systems, agentic AI, and LLM-based evaluation.

I build both research-driven ML algorithms and applied AI systems, including retrieval pipelines, tool-using agents, decentralized learning codebases, and real-time edge-cloud inference systems.

Experience

AI Research Scientist, Knova AI | Jan. 2026 - Present
Build RAG systems, retrieval/reranking pipelines, LLM evaluation workflows, and tool-using customer-support agents.
Research Assistant, INSPIRE Lab, Rutgers University | Feb. 2020 - Jan. 2026
Developed robust decentralized ML algorithms with theoretical convergence analysis and TensorFlow experiments.
Project Researcher, WINLAB/RINGS | May 2022 - May 2025
Worked on real-time edge-cloud inference, federated inference, and active/online learning.
Teaching Assistant, Rutgers ECE | Selected semesters
Supported Digital Electronics and Electronic Devices labs and lecture instruction.

Featured Projects

AcademicRAG Q&A System - retrieval and reranking pipeline that improved correctness by 13% and fairness by 25%, reaching 83% correctness and 76% fairness.
LangGraph Customer-Support Agent - 4-tool ReAct agent with RAG search, account lookup, network-status checks, and human escalation.
Robust Decentralized ML Codebases - first-author codebases for BRIDGE, RESIST, and CUBED-GD with 50-agent TensorFlow experiments.
RINGS Edge-Cloud Inference - real-time edge-cloud inference on the ORBIT testbed with about 100 ms end-to-end latency.

Publications

BRIDGE: Byzantine-Resilient Decentralized Gradient Descent - IEEE TSIPN.
RESIST: Robust Decentralized Learning with Consensus Gradient Descent - TMLR under peer review.
CUBED-GD: Communication-efficient Byzantine-resilient Decentralized Gradient Descent - TMLR submitted.

Page updated

Google Sites

Report abuse