模拟面试提问 Mock Interview
你好,我是买宇翔(英文名 Miles)。有什么问题尽管问我,我会用第一人称回答。
Hi, I'm Yuxiang Mai (Miles). Ask me anything—I'll answer in first person.
金融 AI 产品 FDE Finance AI PM · FDE
金融 AI 产品经理 · CFA 持证人 · 10 年金融 + AI 解决方案落地 Financial AI PM · CFA Charterholder · 10 Years in Finance + AI Solutions
工作经历 Work Experience
Business Analysis Officer Business Analysis Officer
2021.11 — 至今
美国道富(浙江)有限公司 State Street (Zhejiang) Co., Ltd.
AI基金数据监控 Agent — 产品设计 | 2025 — 至今 AI Fund Data Monitoring Agent — Product Design | 2025 – Present
- 从日常BA工作中发现运营团队手工监控与核对基金数据占工时近一半的核心痛点,联合运营部门做根因分析,作为AI解决方案主导自动化立项
- 主导AI数据监控Agent从需求到交付:撰写PRD,提出可解释AI三层处理架构(规则引擎+LLM Agent+人工兜底),确保合规可控、全程可审计;基于Function Calling对接Bloomberg等第三方数据源,按类型化schema调用工具获取实时估值与持仓数据,输出中的每个数字均可溯源到具体工具调用
- 规划三阶段灰度发布策略,一期上线后人工监控工时缩短近40%
- Identified from daily BA work that manual monitoring and verification of fund data consumed nearly half of the operations team's hours; ran root-cause analysis with the operations department and led the automation initiative as an AI solution.
- Drove the AI data-monitoring Agent from requirements to delivery: authored the PRD, proposed an explainable-AI three-tier architecture (rules engine + LLM Agent + human fallback) ensuring compliance and full auditability; integrated Bloomberg and other third-party data sources via Function Calling with typed schemas, so every figure in the output is traceable to a specific tool call.
- Designed a three-phase canary release strategy; phase one cut manual monitoring hours by nearly 40%.
团队 Coaching 与 AI 转型 | 2024 — 至今 Team Coaching & AI Transformation | 2024 – Present
- AI Workflow工作流梳理:推动团队采用自然语言描述→自动生成AS-IS/TO-BE流程图草稿
- AI测试用例生成:推动测试团队基于PRD验收标准,用AI批量生成边界值与异常场景测试用例
- AI组织转型:组织Copilot使用培训,带动团队从工具试用向常态化AI工作流转型
- AI Workflow streamlining: drove the team to adopt natural-language-to-flowchart generation, producing AS-IS/TO-BE flowchart drafts automatically.
- AI test-case generation: guided the QA team to batch-generate boundary and edge-case test cases from PRD acceptance criteria using AI.
- AI organizational transformation: organized Copilot training, shifting the team from tool trial to normalized AI-workflow adoption.
资管数据平台迭代升级 | 2022 — 至今 Asset Management Data Platform — Iteration & Upgrade | 2022 – Present
- 推动CRIMS前台交易系统与SSCD数据报告系统融合,参与多次版本迭代,解决前台交易与后台报告数据割裂问题,覆盖交易、估值与TA(过户登记/投资者申赎)数据流
- 制定UAT测试策略,编写测试用例,拦截10+个高危缺陷;对接美欧市场客户(监管与运营要求各异)需求,通过Jira管理Backlog与Sprint,推动产品交付
- Drove the convergence of the CRIMS front-office trading system with the SSCD data reporting system across multiple version iterations, resolving the disconnect between front-office trading and back-office reporting data, covering trading, valuation, and TA (transfer agency / investor subscription-redemption) data flows.
- Defined the UAT test strategy and authored test cases, intercepting 10+ high-severity defects; gathered requirements from US/EU clients with divergent regulatory and operational requirements, managed Backlog and Sprint via Jira, and drove product delivery.
客户定制化产品敏捷交付 | 2023 — 2025 Client Customized Product Agile Delivery | 2023 – 2025
- 主导Settled Cash Reconciliation Report与Daily Detailed NAV Compare Report等客户定制产品从需求到上线,自动化报告生成节约运营时间约70%
- Led client-customized products such as the Settled Cash Reconciliation Report and Daily Detailed NAV Compare Report from requirements to launch; automated report generation cut operations time by ~70%.
高级基金会计 Senior Fund Accountant
2021.06 — 2021.08
CITCO(加拿大)有限公司 CITCO (Canada) Inc.
- 管理多只私募股权基金的会计核算与资产估值,审查财务数据并编制NAV报告,协调资金催缴及收益分配;因疫情及家庭原因决定回国发展
- Managed accounting and asset valuation for multiple private-equity funds, reviewed financial data and prepared NAV reports, and coordinated capital calls and distributions; returned to China due to pandemic and family reasons.
Operations Associate Operations Associate
2019.07 — 2021.01
SS&C Technologies(加拿大) SS&C Technologies (Canada)
- 与券商及基金经理合作核算交易差异、控制交易风险,使用彭博终端验证场内场外衍生品定价;开发自动化工具将数据导入时间从2小时缩短至15分钟
- Partnered with brokers and fund managers to resolve trade breaks and control trading risk; validated exchange-traded and OTC derivatives pricing using the Bloomberg Terminal; developed automation tooling that cut data-import time from 2 hours to 15 minutes.
运营分析师 Operations Analyst
2017.05 — 2019.05
CITCO (Canada) Inc CITCO (Canada) Inc
- 负责 Transfer Agency(TA,过户登记)运营:投资者申购赎回订单处理、多资产类别基金交易运营与每日现金对账;开发期货佣金对账等自动化工具覆盖数据提取到报表生成,将手工处理时间从8小时缩短至1小时,累计节省0.5个FTE
- Ran Transfer Agency (TA) operations: investor subscription/redemption order processing, multi-asset fund trade operations and daily cash reconciliation; built automation tooling (e.g., futures commission reconciliation) covering data extraction to report generation, cutting manual processing time from 8 hours to 1 hour and saving 0.5 FTE cumulatively.
合伙人-辅导老师(自雇) Partner / Tutor (Self-employed)
2014.09 — 2017.04 全职,2017.05 — 2019.06 兼职
博洋教育 Boyang Education
- 主讲本科及硕士商科金融课程(公司金融、投资组合管理、衍生品等),编写备课方案,培训新辅导老师;组建并运营本地CFA培训项目
- Taught undergraduate and graduate finance courses, authored lesson plans, trained new tutors, and built and ran a local CFA training program.
个人项目 Personal Projects
Lumina Lumina
通用 AI Working Agent(自研 Agent Harness) General-Purpose AI Working Agent (Self-Built Agent Harness)
- 自研 Agent Loop(Electron + FastAPI),通过 MCP 与 Skill 机制实现工具与能力的可插拔扩展,新能力接入不改动核心循环
- Build / Ask / Plan 三档权限模式,高风险操作强制人工确认;对话地图分叉机制支持同一任务多决策路径的调试、对比与回滚
- Self-built Agent Loop (Electron + FastAPI) with pluggable tools and capabilities via MCP and Skill mechanisms, so new capabilities are added without touching the core loop.
- Build / Ask / Plan three-tier permission modes with mandatory human confirmation for high-risk operations; conversation-map branching for debugging, comparing, and rolling back multiple decision paths of the same task.
StockResearch StockResearch
Multi-Agent AI 投研终端(GitHub 100+ ⭐) Multi-Agent AI Research Terminal (GitHub 100+ ⭐)
- LangGraph 编排多 Agent 并行架构:基本面/技术面/情绪面/宏观面独立分析→交叉辩论→裁判 Agent 合成结论,内置金融知识库驱动跨资产研究推理
- 个人/专家双模式共用同一事实层 JSON;BYOK 密钥本地存储、数据不上云;所有结论来源可追溯,数据缺失时显式降级而非让 LLM 编造
- LangGraph-orchestrated multi-Agent parallel architecture: fundamental / technical / sentiment / macro Agents analyze independently → cross-debate → a judge Agent synthesizes the conclusion; a built-in financial knowledge base powers cross-asset reasoning.
- Personal/expert dual modes share one fact-layer JSON; BYOK keys stored locally, no data uploaded to the cloud; every conclusion is source-traceable, with explicit degradation instead of LLM fabrication when data is missing.
NoteAI NoteAI
本地 RAG 知识图谱应用 Local RAG Knowledge-Graph App
- RAG 全链路实践:HyDE 查询扩展 + 向量数据库与结构化知识库混合检索 + Rerank,实现知识消化→知识编译→语义问答闭环
- 数据驱动调优:Top-K 等检索参数根据用户反馈自动调整;Tauri 桌面端 + Python sidecar 架构,数据本地优先
- Full-chain RAG practice: HyDE query expansion + hybrid retrieval across vector database and structured knowledge base + Rerank, enabling a closed loop of knowledge ingestion → compilation → semantic Q&A.
- Data-driven tuning: Top-K and other retrieval parameters auto-adjust based on user feedback; Tauri desktop + Python sidecar architecture, local-first data.
教育背景 Education
金融学 · 硕士 MSc Finance
2013 — 2014
圣玛丽大学 Saint Mary's University
经济学 · 本科 BSc Economics
2008 — 2012
河南工业大学 Henan University of Technology