Agentic Systems
Runtimes, planning, tools, memory and evaluation for agents that act over long horizons.
Runtime Observatory · 2026
YuechengWen
I study how AI agents can reason, collaborate, and remain observable under real-world constraints.
CS Undergraduate · Agent Systems Builder · Chengdu, China
Seeking AI postgraduate study & research opportunities
00 · Identity layer
Academic profile
I am a Computer Science undergraduate at Sichuan University, focused on agentic systems, reliable AI and retrieval. My work connects research questions with deployable systems: from a self-built agent runtime and evaluation harness to safety-aware multi-agent collaboration.
3.85/4.00GPA
91.01/100Weighted average
26/320Cohort rank
7.0IELTS
01 · Academic record
Education & selected coursework
A rigorous foundation in algorithms, systems and machine learning, extended through independent agent engineering and multimodal AI research.
Sichuan University
B.Eng. in Computer Science and Technology
2023 — 2027 (Expected)
- First-Class Individual Scholarship, Sichuan University
- Second Prize (Provincial), National Undergraduate Mathematics Competition
- Third-Class Scholarship, Sichuan University
- Third Prize (Provincial), Mathematical Contest in Modeling
Score
- Algorithms98
- Data Structures94
- Calculus95
- Linear Algebra93
- Operating Systems95
- Computer Networks98 · Project 100
02 · Research vector
Research interests
Centered on agentic systems—and the question of how models can act reliably in complex environments.
Reliable AI
Observability, permissions and evidence-based evaluation for dependable AI applications.
Multi-Agent Collaboration
Event-driven coordination, role specialization and verifiable artifact hand-offs.
Retrieval-Augmented Generation
Hybrid retrieval, routing, reranking and evaluation for grounded reasoning.
AI Safety & Multimodal Reasoning
Risk-aware generation and reasoning across language and visual evidence.
03 · Engineered evidence
Featured systems
Each system is presented through its problem, architecture, personal contribution and evaluation evidence.
01 · Agent Runtime
Yuecheng-Harness
A from-scratch coding-agent runtime with observable tool loops, permission boundaries, context folding, MCP interoperability and a skill evolution workflow.
● Self-directed systems research
02 · Safety-Aware Multi-Agent System
MindBridge
A psychology-oriented multi-agent platform that connects intent routing, knowledge retrieval, risk governance, durable reports and run traces through an event-driven runtime.
● Deployed · evaluation in progress
03 · Retrieval & Learning
EduRAG Agent
An education-focused RAG agent combining hybrid retrieval, document parsing, query routing, mock interviews and learning-plan generation.
● Research prototype
04 · LLM Application Platform
Chat2Excel
A natural-language spreadsheet and database assistant built around schema-aware NL2SQL, asynchronous exports and observable streaming interaction.
● Full-stack system
04 · Research experience
Research experience
ACL 2025 reproduction & extension
MIND · Multimodal Implicit Toxicity Detection
Reproduced and studied a rule-induced, debate-based framework for recognizing implicit toxicity in memes where harm emerges from interactions between text and imagery.
- Implemented dual debater–judge reasoning with self-supervised rule induction.
- Examined rule selection and cross-modal evidence on FHM, HarM and MAMI benchmarks.
FHM: 70% accuracy · 0.60 Macro-F1 in the reproduced setup
05 · Capability map
Technical skills
Agent & LLM Systems
- Agent runtimes
- Multi-agent orchestration
- MCP
- Tool use
- Memory & context engineering
- Evaluation harnesses
Retrieval & ML
- RAG
- BM25
- BGE-M3
- Vector databases
- Reranking
- RAGAS
- CLIP
- ResNet
Engineering
- Python
- C++
- SQL
- Java
- FastAPI
- Spring Boot
- Redis
- Docker
- Caddy
Research Practice
- Experiment design
- Benchmark construction
- Ablation thinking
- Trace analysis
- Technical writing
06 · Outside the runtime
Field Notes
Photography is how I practice patient observation beyond code—light, structure and small signals in changing environments.
