Di Zhang Updates Di Zhang is a PhD candidate at Fudan University working on LLM reasoning, scientific intelligence, agentic learning, and foundation models for scientific discovery. https://trotsky1997.github.io/updates/ 2026-06-06T00:00:00+08:00 Di Zhang di.zhang@ustc.edu Agent-Own Loop Harness https://trotsky1997.github.io/updates/#agent-own-loop-harness 2026-06-06T00:00:00+08:00 Blog entry on turning OpenSpec from a human-attended workflow into an agent-owned SDD loop. delta-mem: Efficient Online Memory for Large Language Models https://trotsky1997.github.io/updates/#delta-mem-online-memory 2026-06-06T00:00:00+08:00 Publication entry added for delta-mem, an efficient online memory approach for large language models. MinT: Managed Infrastructure for Training and Serving Millions of LLMs https://trotsky1997.github.io/updates/#mint-managed-infrastructure 2026-06-06T00:00:00+08:00 Publication entry added for MinT, a managed infrastructure system for training and serving large numbers of LLMs. Harness Engineering https://trotsky1997.github.io/updates/#harness-engineering-shell-for-the-ghost 2026-04-04T00:00:00+08:00 Blog entry framing harnesses as the operating shell around agentic systems. Triquetra for LoRAs https://trotsky1997.github.io/updates/#triquetra-lora-rank-learning-rate 2026-03-24T00:00:00+08:00 Blog entry on how LoRA rank, alpha, and learning rate move together when update scale changes.