M.S. student in Applied Mathematics · Fudan University
English · 简体中文
I am an Applied Mathematics M.S. student at Fudan University. My current interests include Scientific Machine Learning (Scientific ML), neural operators and PDEs, AI for Science & Mathematics (AI4Science / AI4Math), and practical AI systems.
Alongside research, I maintain public projects and contribute upstream to open-source software, with current work spanning AI agents and agent reliability, research-intelligence tooling, reproducible research workflows, scientific computing, and developer tooling.
| Area | Exploring |
|---|---|
| Scientific ML | Neural operators, PDEs, and scientific machine learning for physical systems |
| Agent Engineering | AI agents, agent reliability, research agents, and developer tooling |
| Open Source | Public projects, upstream contributions, and reproducible research workflows |
|
A timeline-first, evidence-backed archive of major milestones in AI for Mathematics. A continuously maintained bilingual project with a public site, explicit editorial and verification methodology, machine-readable data, and citable releases.
|
A curated, evidence-aware research-intelligence project for AI for Mathematics. Independently maintained and derived from the AIHOT open-source framework; combines an explicit selection policy, structured AI4Math data, deterministic validation, and a static public site. v0.1.0 is the first citable editorial baseline.
|
|
Agent-orchestrated, reproducible engineering simulation workflows with explicit physics validation. A public lab for learning and building trustworthy automation around Ansys-based simulation workflows, with solver-derived evidence and structured validation.
|
Compare the workspace instruction surfaces seen by DeepSeek Harness, Codex, and Claude Code. A small developer tool that makes cross-agent instruction discovery differences visible, with explicit observed/predicted evidence semantics. The v0.1 release is available on npm and GitHub Releases.
|
|
An auditable testbench for a typed probabilistic decision primitive. Measures TypeSafe Jev through append-only evidence and preregistered experiments, with a frozen evidence release and an explicit null result rather than a polished success narrative.
|
Reusable engineering preferences, decision rules, and project conventions for AI-assisted software development. Designed as secondary context for coding agents, with project-local instructions taking precedence and selected public examples showing where the preferences are used in practice.
|
- DeepMathLLM / Creative-Intelligence — merged work on deterministic regression/integration testing, resumable-workflow provenance, and lifecycle integrity; further verification/controller hardening remains under review.
- DeepMathLLM / Moonshine — open PR for crash-safe tool execution journaling and interrupted-turn recovery.
- KKKKhazix / AIHOT — merged work on pairwise event-relation evaluation, SelectBench semantics, and MCP smoke-contract checks; PR #56 for weekly-report API exposure is under review.
- OpenHands / software-agent-sdk — merged PR #5029 normalizing LLM usage telemetry through a typed adapter.
- Scientific Machine Learning
- Neural Operators & Partial Differential Equations (PDEs)
- AI for Science & Mathematics (AI4Science / AI4Math)
- Learning-based methods for physical systems
Open to conversations around research, open-source projects, and AI engineering.
