Skip to content

Kevin 拾玖秋AI Application & Agent Engineer

Available for work

Shanghai GMT+8

KevinAI Application& Agent Engineer

Building useful AI systems,not AI demos.

Building practical AI systems with Agents, RAG, workflows and local-first tools.

Role
AI Application & Agent Engineer
Focus
Agent · RAG · Workflow
Proof
2059 tests passed · JobMate
Based in
Shanghai, China

Selected Work

01—03 / 5 projects

01

JobMate

AI Job Application Copilot

A local-first AI workflow for turning resumes and live job descriptions into actionable application drafts.

Type
AI Product · Browser Extension
Stack
Chrome MV3 · Local-first · LLM · Python
Proof
2059 tests passed
Read case study
02

Commissioning Knowledge Retrieval & Troubleshooting System (RAG)

设备调试​知识检索​与​故障辅助排查​系统(RAG)

A RAG system for looking up commissioning documentation and supporting new-hire training: four-way hybrid retrieval plus rerank, answers that must cite source evidence, and a refusal or follow-up question instead of a made-up answer when evidence falls short.

Type
Enterprise Knowledge Retrieval · RAG
Stack
RAG · Hybrid Retrieval · Rerank · FastAPI
Proof
0.993 Recall@5 · 2200 tests passed · 89% line coverage
Read case study
03

Construction Plan AI Generation & Review System

建筑施工方案 AI 生成与审核系统

A multi-stage AI system for construction-plan generation and review, combining retrieval, deterministic tools and agent workflow orchestration.

Type
Multi-Agent System
Stack
Multi-Agent · LangGraph · RAG · Tool Calling
Proof
0.60 Hit@1 · 1.00 Hit@3 · 0.77 MRR · 7 tests passed
Read case study

Open Source

Open Source

Code I maintain in public, and fixes contributed back upstream.
  1. 01Own project

    job-application-assistant

    AI Job Application Agent Skill

    An open-source job-application Agent Skill for Claude and Codex, covering research, application preparation, ATS assistance and tracking.

    Python · Claude Skill · Codex · Playwright · Human-in-the-loop · Local-first

  2. 02Upstream contribution

    volcengine/OpenViking

    Semantic Parent Refresh Namespace Boundary Fix

    Fixed semantic parent refresh crossing into invalid namespace roots and added regression coverage for valid and invalid semantic-root boundaries.

    Repositorygithub.com/volcengine/OpenViking(opens in a new tab)

Approach

I build AI applications around real workflows, not isolated model demos.

Engineering principles

  1. 01

    Problem Before Model

    Solve the business problem first. Choose the model after.

  2. 02

    Workflow Before Agent

    If a workflow can solve it, it doesn't need an agent. Autonomy is a cost, not a feature.

  3. 03

    Evaluate Retrieval

    RAG is not a vector database with a prompt on top. Retrieval has to be measured.

About

From Architecture
to AI Engineering.

Architecture taught me how to structure complex problems. AI engineering gave me a new medium to build solutions.
  1. 2023

    Started exploring Generative AI & LLMs.

  2. 2024

    AI application development & training.

  3. 2025

    AI development in production-oriented scenarios.

  4. 2026

    Agent Engineering / RAG Evaluation / AI Coding / Open Source.

Currently Exploring

  • Agent Architecture

    Agent Loop / State / Executor / Tools

  • RAG Evaluation

    Hit@K / MRR / Retrieval Quality / Ranking Quality

  • Local-first AI

    Privacy-first AI product architecture

  • AI Coding Agents

    Claude Code / Codex / Agentic Software Engineering