7-Role Battle-Tested Playbook
同一周可能上午面 FDE、下午面 FDPM、明天面 Operations 或 Consulting——用这份速查卡快速切换“面试官真正想听到的思维模式与30秒开场白”。
Own the outcome from messy client floor to running production code. Ship value in Week 2 via narrow vertical slices while keeping platform boundaries clean.
以交付结果倒推架构:能用确定性代码解决的绝不交给大模型,先用最小垂直切片跑通生产闭环并接入HITL,再逐步抽象扩展。
- 1. Clarify business SLA, security boundary & dirty legacy interfaces (ERP/PDF/VPN)
- 2. Slice 1 high-ROI workflow into a 30-60-90 rollout (Shadow -> Read-Only -> Write-Back)
- 3. Separate deterministic rules from probabilistic AI + enforce confidence-gated HITL
- 4. Isolate bespoke client glue in adapters so Core Platform stays clean
- • Time-to-First-Value (TTFV)
- • End-to-End Accuracy vs. Recall on Critical Errors
- • Human Override / Fallback Rate
- • P95 Latency & Integration Uptime
- • Core Reusability Ratio (%)
- ✕ Designing a 6-month ideal data lake before shipping anything to real users
- ✕ Letting LLMs do arithmetic, exact joins, or unverified ERP mutations
- ✕ Never checking how client operators actually work day-to-day
"Before drawing boxes, I want to lock down 3 constraints: who the Day-1 operator is, what legacy systems we must read/write safely, and what accuracy bar unlocks a Week-2 shadow pilot..."
“Same Prompt, 7 Lenses” Cross-Role Comparator
Shared Scenario: "An enterprise client wants an AI agent to automate invoice exception handling across 5 regional ERPs."
同一道场景题,7 个岗位在面试前 90 秒的切入视角对比
Audit the 5 ERP APIs/batch dumps, pick 1 region + top 2 exception types for Week-2 shadow mode, and isolate deterministic matching from LLM PDF extraction.
Canonical Event Schema + ERP Adapter Diagram + Shadow Log Pipeline
Extraction/Match Precision & Zero Silent Bad ERP Writes
Air-gapped SAP permissions, duplicate write-backs, dirty OCR tables
Identify the economic buyer (VP Finance/AP), lock pilot ROI target ($ recovered/FTE hours freed), and separate the 80% reusable AP Engine from 20% regional tax rules.
Pilot ROI Charter + 80/20 Platform vs. Bespoke Boundary Spec
Pilot-to-Enterprise ARR Conversion & Verified Payback Months
Building 5 bespoke regional forks that Core Eng refuses to maintain
Frame total value pool (early-payment discounts captured + duplicate leak recovery + labor efficiency vs. build/run TCO) and rank which 2 regions have fastest ROI.
MECE Value-Pool Tree + 3-Year NPV Business Case + CEO Pyramid Deck
Net EBITDA Impact ($M) & 12-Month Payback Hurdle
Overestimating soft headcount savings when regional ERP complexity eats margin