I am a System Software Development Engineer at Intel 🏢 in Beijing, with over 2.5 years of experience. My research and work primarily focus on LLM inference optimization and efficient AI 🤖, including model compression (pruning, sparsity, quantization), inference acceleration⚡, and efficient hybrid GPU-CPU engine for Agentic AI. My work bridges the gap between high-level algorithmic research and low-level system efficiency.

Research Interests 🚩

• Natural language processing (NLP), large language models (LLM).
• LLM inference optimization, model compression (pruning, sparsity, and quantization), neural architecture search (NAS).
• LLM fine-tuning, parameter-efficient fine-tuning (PEFT), LoRA.
• LLM serving engine, hybrid GPU-CPU inference.
• Agentic AI, KV-cache management, agent memory management, retrieval-augmented generation (RAG), test-time compute, agent skills.

I am always open to discussions and potential collaborations. Please feel free to reach out! 🤝

Experience 💼

• System Software Development Engineer @ Intel
  Jul. 2023 - Present · Beijing, China

• System Software Development Engineer (Intern) @ Intel
  May 2022 - Jun. 2023 · Beijing, China

• M.S. in Computer Science and Technology @ Guangdong University of Technology
  Sep. 2020 - Jun. 2023 · Guangzhou, China
  Supervisor: Prof. Ruichu Cai @ DMIR Lab
  Research: NLP, Text-to-SQL
  Recommended for Admission without Examination

• B.S. in Software Engineering @ Guangdong University of Technology
  Sep. 2016 - Jun. 2020 · Guangzhou, China

Selected Publications 📚

For a full list, please refer to my Google Scholar.

• RTTC: Reward-Guided Collaborative Test-Time Compute
  J. Pablo Muñoz*, Jinjie Yuan* (Co-first Author)
  EMNLP 2025 Findings · Paper

• Mamba-Shedder: Post-Transformer Compression for Efficient Selective Structured State Space Models
  J. Pablo Muñoz*, Jinjie Yuan* (Co-first Author), Nilesh Jain
  NAACL 2025 (Oral) · Paper · Code

• Low-Rank Adapters Meet Neural Architecture Search for LLM Compression
  J. Pablo Muñoz, Jinjie Yuan, Nilesh Jain
  AAAI 2025 Workshop on Connecting Low-rank Representations in AI · Paper

• SQFT: Low-cost Model Adaptation in Low-precision Sparse Foundation Models
  J. Pablo Muñoz*, Jinjie Yuan* (Co-first Author), Nilesh Jain
  EMNLP 2024 Findings · Paper · Code

• Shears: Unstructured Sparsity with Neural Low-rank Adapter Search
  J. Pablo Muñoz*, Jinjie Yuan* (Co-first Author), Nilesh Jain
  NAACL 2024 · Paper · Code

• LoNAS: Elastic Low-rank Adapters for Efficient Large Language Models
  J. Pablo Muñoz*, Jinjie Yuan* (Co-first Author), Yi Zheng, Nilesh Jain
  LREC-COLING 2024 · Paper · Code

• SADGA: Structure-Aware Dual Graph Aggregation Network for Text-to-SQL
  Ruichu Cai, Jinjie Yuan (Advisor 1st, Student 2nd), Boyan Xu, Zhifeng Hao
  NeurIPS 2021 · Paper · Code

Patents 📝

• Methods and apparatus for enabling efficient fine-tuning on unstructured sparse and low-precision large pre-trained foundation models
  J. Pablo Muñoz, Jinjie Yuan, Nilesh Jain
  US Patent App. 18/935,223 (2025)