Resume / Experience

简历

Resume / Experience

技术栈

FastAPIReactAgent LoopTool CallingRAGLangGraphChromaDBBM25RRFpytestJavaRedisMQ

简介

Computer Technology master's student at Zhejiang Sci-Tech University, targeting AI application development and backend engineering internships. Main projects include NexaRAG and RunLens.

Education

Zhejiang Sci-Tech University · Computer Technology M.S.

Focus

Agent / RAG / Backend Engineering

亮点

Hands-on Agent application experience covering Agent Loops, tool calling, context management, planning and state progression, permissions, failure recovery, and result validation.
Hands-on RAG system experience covering ingestion, structured chunking, embedding + BM25 hybrid retrieval, query rewriting, reranking, context engineering, citations, and evaluation.
Skilled in using Claude Code / Codex for code understanding, feature development, debugging, test completion, and refactoring, with explicit result verification.
Used Agent systems such as OpenClaw / Hermes and understand tool scheduling, context passing, execution loops, state management, and result write-back.
Java backend project experience with layered application design, Redis caching, asynchronous processing with MQ, and service decoupling.

Project Experience

2026.03 - Present

NexaRAG: RAG Customer Support System for 3C Digital Products

Built an intelligent support system for product consultation, troubleshooting, recommendations, competitor comparison, small-talk routing, and image Q&A, with an engineering focus on retrieval quality, context construction, and grounded answers.

Structured Chunking and Metadata: split documents along heading, paragraph, and procedural boundaries, preserving source, product, type, and version for filtering, citations, and updates.
Hybrid Retrieval and Context Engineering: combined query rewriting, vector retrieval, BM25, RRF, and reranking, then applied deduplication and context budgets.
LangGraph State Orchestration: managed state passing and conditional routing across intent recognition, small talk, product Q&A, troubleshooting, recommendations, and image Q&A.
Answer Validation and Evidence Governance: verified citations, refused unsupported answers, and surfaced uncertainty when retrieved evidence conflicted.
End-to-end Storage and Golden Sets: persisted rewriting, retrieval, ranking, context, citations, and answers for continuous retrieval and response evaluation.

GitHub

2025.10 - 2026.01

RunLens: Verifiable, Observable, Plugin-extensible Local Agent Runtime

Designed and implemented a synchronous, single-process, file-backed Agent Runtime with shared execution contracts that constrain model actions through permissions, state, recovery, and delivery conditions.

Plugin-based Agent Runtime: abstracted model loops, tools, permissions, state machines, recovery, and validation so businesses declare only inputs, skills, plans, tools, and delivery contracts.
Recoverable and Terminable Agent Loop: implemented model calls, ToolCalls, observation write-back, budgets, repeated-call detection, provider retries, and termination policies.
Tool Contracts and Permission Boundaries: passed ToolCalls through lookup, schemas, permissions, hooks, handlers, and output validation, with filesystem path allowlists.
Artifact Contracts and Evidence Acceptance: validated required files, JSON structures, Markdown sections, and evidence references before a run could complete.
Event-driven Observability: recorded runtime facts on one timeline and derived traces, metrics, and manifests for debugging, statistics, and presentation.

GitHub