AI 长期知识图谱与复习系统
当前仓库已经不是单一学习计划, 而是一个多阶段能力资产库:
AI Long-Term Knowledge Graph and Review System
定位: 把已有 Web3、金融零售、架构、LLM、AIPA、ABPA、AI Foundations、职业资产扩展连接成一张长期可复习、可追问、可转作品集的知识图谱。 原则: 不删除旧内容, 不把旧计划判定为废弃。旧资产是长期素材库, 新系统负责导航、复习、更新和证据转换。 目标: 未来 12-18 个月持续增强 AI Solutions Architect / AI Business Architect / AI PM / AI BA 能力。
1. 为什么需要这张长期图谱
当前仓库已经不是单一学习计划, 而是一个多阶段能力资产库:
- Web3 90 天和实战计划: 产品机制、链上数据、协议分析、Tokenomics、治理。
- 架构 120+ 和专家计划: 金融零售业务架构、支付、核心银行、供应链、DDD、微服务、云原生。
- DSDB / LLM / AIPA: 分布式系统、数据库、LLM 原理、Agent、Eval、MCP、AML Copilot、AI-native architecture。
- ABPA-180: BA/PM/架构融合, 从问题定义、流程、需求、eval、治理到 adoption。
- AI Foundations: Transformer、RAG、Agent、RLHF、CoT、LoRA、推理优化、LLM-as-Judge。
- 职业资产扩展: 角色能力矩阵、金融零售 AI case portfolio、架构图谱、面试和作品集。
如果没有知识图谱, 风险是:
- 学了很多, 但面试时讲不成故事。
- 写了很多, 但不知道哪些能证明能力。
- 旧内容越来越多, 但复习路径不清楚。
- 新 AI 方向不断变化, 但旧资产无法复用。
- 能力增长停留在“读过”, 没变成“能判断、能设计、能落地、能复盘”。
这张图谱的目的不是再增加一条学习负担, 而是把所有已有资产组织成:
Knowledge Node -> Capability -> Artifact -> Evidence -> Interview Story -> Next Review
2. Source Anchors
本图谱引用的标准和方法只作为学习与组织框架, 不构成法律、审计、采购或合规意见。
| Anchor | Link | 用法 |
|---|---|---|
| NIST AI RMF | https://www.nist.gov/itl/ai-risk-management-framework | 将 AI 风险转成 govern/map/measure/manage 的复习与控制维度 |
| NIST AI 600-1 GenAI Profile | https://www.nist.gov/itl/ai-risk-management-framework | 为 GenAI 风险建立额外复习维度: hallucination, data privacy, misuse, evaluation |
| EU AI Act | https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng | 用 risk-based thinking 组织高风险 AI 场景的证据和透明度要求 |
| ISO/IEC 42001 | https://www.iso.org/standard/42001 | 用 AI management system 思路组织责任、生命周期和持续改进 |
| OWASP LLM Top 10 | https://owasp.org/www-project-top-10-for-large-language-model-applications/ | 用 LLM 风险类别建立安全复习清单 |
| TOGAF | https://www.opengroup.org/togaf | 用企业架构层次组织业务、数据、应用、技术、治理和路线图 |
| OMG BPMN | https://www.omg.org/spec/BPMN/ | 用流程图谱连接 BA 需求、工作流、异常和控制 |
| BIAN | https://bian.org/deliverables/service-landscape/ | 用银行服务域把金融场景拆成能力和系统边界 |
3. 总体知识图谱
MomoWeb3 / AI Learning Asset Graph
├── Domain Depth
│ ├── Financial Retail: AML, KYC, lending, fraud, payments, wealth, service ops
│ ├── Web3 Background: DeFi, RWA, wallet, governance, tokenomics, security
│ └── Business Architecture: capability, value stream, domain, operating model
├── AI Foundations
│ ├── Transformer and LLM mechanics
│ ├── RAG and knowledge architecture
│ ├── Agent and tool-use architecture
│ ├── Alignment and feedback
│ ├── Reasoning and test-time compute
│ ├── Adaptation: LoRA / PEFT / fine-tuning
│ ├── Inference optimization
│ └── Evaluation: LLM-as-Judge / EvalOps
├── AI Delivery System
│ ├── Problem framing
│ ├── BPMN workflow discovery
│ ├── Requirements-to-Eval
│ ├── Data readiness
│ ├── Architecture ADR
│ ├── AI control pack
│ ├── Operating model and RACI
│ ├── Adoption dashboard
│ └── Business case
├── Enterprise AI Architecture
│ ├── C4 context/container/component
│ ├── Model gateway
│ ├── RAG platform
│ ├── Agent runtime and tool gateway
│ ├── Eval platform
│ ├── Observability and runbook
│ ├── Governance and risk operations
│ └── Vendor/build-buy strategy
└── Career Evidence
├── Role competency matrix
├── Case portfolio
├── Diagram playbook
├── Interview bank
├── Portfolio evidence map
└── Flagship capstone stories
4. 六大能力域
Domain A: AI Foundations
目标: 能解释底层机制, 并把机制转成产品和架构取舍。
| Node | 主要资料 | 要能回答的问题 | 证据资产 |
|---|---|---|---|
| Transformer | docs/ai-foundations/papers/01-attention-is-all-you-need.md | 为什么 LLM 能建模上下文? 为什么会有成本和长上下文限制? | Transformer one-pager, C4 model gateway note |
| RAG | docs/ai-foundations/papers/02-retrieval-augmented-generation.md | 什么时候用 RAG, 什么时候不用? RAG 为什么是知识治理? | RAG ADR, data flow, citation eval |
| Agent / Tool Use | docs/ai-foundations/papers/03-react-toolformer-agent-foundations.md | Agent 为什么不是聊天框? 工具权限如何控制? | Agent loop diagram, tool risk matrix |
| RLHF / Alignment | docs/ai-foundations/papers/04-instructgpt-rlhf-alignment.md | 为什么对齐不能替代外部控制? | alignment risk memo, refusal/escalation policy |
| CoT / Reasoning | docs/ai-foundations/papers/05-chain-of-thought-self-consistency.md | 推理质量、成本、解释和证据如何平衡? | reasoning eval set, explanation policy |
| LoRA / PEFT | docs/ai-foundations/papers/06-lora-peft-adaptation.md | 什么时候微调, 什么时候 RAG? | adaptation decision matrix, adapter governance plan |
| Inference Optimization | docs/ai-foundations/papers/07-inference-optimization-kv-cache-flashattention-speculative.md | 成本/延迟/SLO 如何影响 AI 产品? | latency budget, model routing ADR |
| LLM-as-Judge / Eval | docs/ai-foundations/papers/08-llm-as-judge-evaluation.md | 如何评估开放式 AI 输出? | eval rubric, golden set, release gate |
复习重点:
- 每篇论文都要转成一张图。
- 每个机制都要给出一个金融零售案例。
- 每个技术选择都要写出反转条件。
Domain B: AI BA / PM Delivery
目标: 能把“我们想做 AI”变成真实业务问题、流程、需求、验收和 adoption。
| Node | 主要资料 | 要能回答的问题 | 证据资产 |
|---|---|---|---|
| Opportunity Framing | docs/abpa/templates/01-ai-opportunity-canvas.md | 这个问题值得做 AI 吗? | AI opportunity canvas |
| Stakeholder Evidence | docs/abpa/templates/02-stakeholder-evidence-map.md | 谁有 pain、power、evidence、veto? | stakeholder map |
| Process Modeling | docs/abpa/templates/03-bpmn-pain-metrics.md | 当前流程在哪里慢、错、返工、不可控? | AS-IS / TO-BE BPMN |
| Requirements-to-Eval | docs/abpa/templates/04-requirements-to-eval-matrix.md | 需求如何变成测试样本和阈值? | eval matrix |
| Control Design | docs/abpa/templates/05-ai-control-pack.md | 风险如何预防、发现、纠正? | AI control pack |
| Executive Decision | docs/abpa/templates/06-executive-decision-memo.md | 现在应该批准什么? | decision memo |
| Data Readiness | docs/abpa/templates/07-data-readiness-pack.md | 数据是否足够支撑模型和审计? | data readiness pack |
| Architecture ADR | docs/abpa/templates/08-ai-architecture-adr-set.md | 为什么选择这个架构? | ADR set |
| Operating Model | docs/abpa/templates/09-operating-model-raci.md | 上线后谁负责? | RACI |
| Adoption Dashboard | docs/abpa/templates/10-adoption-dashboard.md | 用户是否真的改变工作方式? | adoption dashboard |
| Business Case | docs/abpa/templates/11-business-case.md | 投资是否值得? | business case |
| Portfolio Evidence | docs/abpa/templates/12-portfolio-evidence-map.md | 如何证明能力? | portfolio evidence map |
复习重点:
- 不只背模板, 每月至少填 1 个真实 case。
- 每个 case 必须包含 baseline、eval、control、ROI。
- 面试回答必须引用 artifact, 不只讲观点。
Domain C: Financial Retail AI Cases
目标: 把 10 年金融零售经验转成 AI-native case portfolio。
| Case | 主要资料 | 适合证明的能力 |
|---|---|---|
| AML Investigation Copilot | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md, docs/abpa/capstone-aml/ | 高风险场景、HITL、证据、审计、EvalOps |
| KYC Remediation | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | 数据质量、客户触达、流程闭环 |
| Customer Service RAG | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | RAG、知识治理、adoption |
| Payments Exception Agent | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | Agent tool-use、sequence、异常处理 |
| Lending Assistant | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | 高风险 AI、公平性、政策解释 |
| Fraud Operations Copilot | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | 实时风险、行动边界、误报/漏报 |
| Wealth Compliance Guardrail | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | 适当性、持牌责任、拒答升级 |
| Regulatory Impact Analysis | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | 法规到能力/流程/系统影响映射 |
复习重点:
- 每个 case 要能用 BA、PM、Architect 三种口径讲。
- 每个 case 要能画至少 3 张图: BPMN, C4, Eval/Risk。
- 每个 case 要有一个 weak answer to avoid。
Domain D: Enterprise AI Architecture
目标: 能把 AI 系统设计成可运行、可管、可评估、可审计的企业能力。
| Node | 主要资料 | 要能回答的问题 | 证据资产 |
|---|---|---|---|
| Diagram Playbook | docs/AI_ARCHITECTURE_DIAGRAM_PLAYBOOK.md | 哪个问题该画哪张图? | diagram pack |
| Agentic Enterprise Architecture | docs/AGENTIC_ENTERPRISE_ARCHITECTURE_90_PLAN.md | Agentic system 如何进入企业架构? | 90-day architecture plan |
| Governance / EvalOps / RiskOps | docs/AI_GOVERNANCE_EVALOPS_RISK_90_PLAN.md | 如何建立上线门禁和风险运营? | governance pack |
| Vendor / Build-Buy | docs/AIPA_LONGFORM_6_BUILD_VS_BUY.md and new vendor playbook | 何时买、建、合作、混合? | build-vs-buy matrix |
| Observability | AIPA notes and diagram playbook | 如何发现 AI 失败? | dashboard/runbook |
| Operating Model | ABPA template 09 | 谁维护 prompt、eval、policy、model? | RACI |
复习重点:
- 架构图必须包含数据、控制、eval、审计、运营。
- 每个 ADR 要有 rejected options。
- 每个系统都要说清楚 failure modes 和 rollback。
Domain E: Governance / Risk / Compliance
目标: 能把 AI 风险讲成产品、架构和运营控制, 而不是只写合规口号。
| Risk/Control Area | Source anchor | 复习问题 | 证据资产 |
|---|---|---|---|
| AI risk management | NIST AI RMF | 该场景的 Govern/Map/Measure/Manage 如何落地? | risk register |
| GenAI risks | NIST AI 600-1, OWASP LLM Top 10 | hallucination、prompt injection、data leakage 如何控制? | threat model |
| High-risk AI | EU AI Act | 是否涉及高风险系统逻辑、透明度、人类监督? | risk classification memo |
| AI management system | ISO/IEC 42001 | policy、owner、lifecycle、continual improvement 如何设计? | operating model |
| Financial control | Domain policy | 哪些动作必须人审? 哪些必须审计? | control pack |
复习重点:
- 不把 prompt 当控制。
- 不把模型解释当证据。
- 高风险场景必须有 HITL、audit、release gate、incident process。
Domain F: Career Evidence and Interview
目标: 把能力转成可被招聘方检查的证据。
| Node | 主要资料 | 产出 |
|---|---|---|
| Role Matrix | docs/AI_ROLE_COMPETENCY_MATRIX_2026.md | role gap map |
| Interview Bank | docs/abpa/interview/AI_BA_PM_ARCHITECT_INTERVIEW_BANK.md | 题库和回答 |
| Portfolio Evidence Map | docs/abpa/templates/12-portfolio-evidence-map.md | claim-to-evidence |
| Case Portfolio | docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | flagship story |
| Diagram Playbook | docs/AI_ARCHITECTURE_DIAGRAM_PLAYBOOK.md | whiteboard story |
| Master Portfolio | docs/MASTER_PORTFOLIO.md | 历史资产和转型故事 |
复习重点:
- 每个 claim 至少两条 evidence。
- 每个 story 要有 30 秒、2 分钟、深挖版本。
- 避免只说“我学过”, 要说“我做了哪张图、哪份 eval、哪个 control、哪个 business case”。
5. Crosswalk: 旧资产如何复用到新 AI 方向
Web3 资产
旧 Web3 内容不是废弃, 而是转成以下能力:
| 旧资产 | 新用法 |
|---|---|
| DeFi 协议分析 | 训练机制设计、激励、风险、开放生态产品判断 |
| Tokenomics | 训练 incentive design, adoption, governance thinking |
| DAO 治理 | 迁移到 AI governance, decision rights, community/user trust |
| 链上数据/Dune | 迁移到 AI product analytics, EvalOps dashboard, adoption metrics |
| Web3 安全 | 迁移到 Agent tool risk, prompt injection, permission boundary |
| RWA/合规 | 迁移到金融 AI 高风险场景和监管证据 |
推荐复习动作:
- 每月选 1 篇旧 Web3 笔记, 增加一段 “AI BA/PM/Architect reuse note”。
- 把协议分析结构迁移到 AI vendor / AI platform 分析。
- 把 governance 经验迁移到 AI control pack。
金融零售架构资产
| 旧资产 | 新用法 |
|---|---|
| 支付系统 | Payments exception agent, settlement ops copilot |
| 核心银行 | customer/account data boundaries, data readiness |
| 信贷/风控 | lending assistant, fair lending, policy explanation |
| 供应链/零售 | operations copilot, demand planning AI, exception handling |
| DDD/微服务 | domain modeling, service boundary, C4 architecture |
| TOGAF/架构治理 | enterprise AI capability and roadmap |
推荐复习动作:
- 每季度选 1 个金融域做 AI case portfolio。
- 每个域都要画 capability map + BPMN + C4。
- 用 BIAN service domain 练银行能力边界。
LLM/AIPA 资产
| 旧资产 | 新用法 |
|---|---|
| Agent eval | EvalOps / release gate |
| MCP / tool gateway | Agentic architecture and tool risk |
| AML Copilot | Flagship capstone |
| Observability | AI production operations |
| Build-vs-buy longform | Vendor due diligence |
| AI-native architecture | Interview and portfolio core |
推荐复习动作:
- 每个 AIPA 作品都补一个 BA/PM framing。
- 每个工程资产都补一个 executive memo。
- 每个 eval 资产都补一个 risk/control mapping。
6. 复习节奏
每日 30 分钟轻复习
只做一件小事:
- 读一页经典论文解读。
- 画一个小图。
- 改进一个 interview answer。
- 给旧笔记加一个 AI reuse note。
- 写一个 eval case。
- 给一个 case 补一个 risk control。
每周 3 小时深复习
建议固定节奏:
| Day | 动作 |
|---|---|
| Monday | 选一个 case 和一个角色视角 |
| Tuesday | 复习相关论文/架构机制 |
| Wednesday | 画 BPMN/C4/RAG/Agent/Eval 图 |
| Thursday | 写 Requirements-to-Eval 或 AI Control Pack |
| Friday | 写 30 秒和 2 分钟面试回答 |
| Weekend | 更新 Portfolio Evidence Map |
每月 1 个证据包
每月只要求产出一个证据包, 但要完整:
Case summary
+ Opportunity Canvas
+ BPMN
+ Architecture diagram
+ Requirements-to-Eval
+ Control Pack
+ Business Case or Adoption Dashboard
+ Interview Story
每季度 1 个旗舰故事
每季度选择一个旗舰故事打磨到可展示:
| Quarter Theme | Suggested Flagship |
|---|---|
| Q1 | AML Investigation Copilot |
| Q2 | Customer Service RAG / Product Knowledge RAG |
| Q3 | Payments Exception Agent |
| Q4 | Lending Assistant / Wealth Compliance Guardrail |
7. 12-Month Learning and Evidence Roadmap
Month 1: AI Foundations consolidation
目标:
- 能用自己的话解释 8 篇 AI Foundations。
- 每篇有一张图。
- 每篇有一个金融零售映射。
证据:
AI Foundations Concept Map v1- 8 个 30 秒回答
- 2 个 ADR
Month 2: AML flagship case
目标:
- 以 AML Copilot 打完整 ABPA artifact loop。
证据:
- Opportunity Canvas
- BPMN
- RAG/Agent architecture
- Eval matrix
- Control pack
- Executive memo
Month 3: Customer service / product knowledge RAG
目标:
- 训练低风险高采用场景。
- 强化 RAG、知识治理、adoption。
证据:
- RAG ADR
- knowledge ingestion diagram
- citation eval
- adoption dashboard
Month 4: Payments exception agent
目标:
- 训练 tool-use、workflow automation、human approval。
证据:
- sequence diagram
- tool risk matrix
- action approval policy
- incident runbook
Month 5: Lending assistant
目标:
- 训练高风险 AI、解释、政策边界、公平性。
证据:
- fair lending control pack
- adverse action explanation eval
- high-risk human oversight memo
Month 6: Vendor / Build-vs-Buy
目标:
- 能做 AI vendor due diligence。
- 能解释何时买、建、合作、混合。
证据:
- vendor scorecard
- build-vs-buy matrix
- pilot criteria
- risk acceptance memo
Month 7: Enterprise AI governance
目标:
- 把 NIST/EU/ISO/OWASP 转成企业 AI operating model。
证据:
- AI governance model
- release gate framework
- risk tiering
- model/prompt/knowledge change control
Month 8: EvalOps platform story
目标:
- 能讲 AI quality loop。
证据:
- golden set taxonomy
- LLM-as-Judge rubric
- release dashboard
- incident review template
Month 9: Architecture portfolio
目标:
- 把 3 个 case 画成一致的 architecture pack。
证据:
- capability map
- C4 context/container
- RAG/Agent diagram
- risk/control diagram
- cost/latency model
Month 10: Business case and ROI
目标:
- 能把 AI 从技术方案讲成投资决策。
证据:
- 3 个 business case
- cost model
- sensitivity analysis
- funding gate memo
Month 11: Interview preparation
目标:
- 每个 flagship story 都有 30 秒、2 分钟、deep dive。
证据:
- interview proof pack
- role-specific answer bank
- objection handling sheet
Month 12: Portfolio packaging
目标:
- 汇总为可投递 portfolio。
证据:
- master AI portfolio README
- 3-5 case studies
- architecture diagram gallery
- eval/control appendix
- resume bullets
8. Spaced Repetition System
复习粒度
不要按“文件”复习, 要按“能力节点”复习:
- Concept: Transformer, RAG, Agent, Eval, LoRA。
- Artifact: BPMN, ADR, Control Pack, Business Case。
- Case: AML, KYC, Payments, Lending。
- Story: 30 秒 / 2 分钟 / deep dive。
- Source anchor: NIST, EU AI Act, ISO, OWASP, TOGAF, BPMN, BIAN。
复习间隔
| 时间 | 动作 |
|---|---|
| D+1 | 用 5 句话复述 |
| D+7 | 画图或填一个模板 |
| D+30 | 用一个新 case 应用 |
| D+90 | 转成面试回答 |
| D+180 | 放进作品集或淘汰 |
Flashcard 类型
| 类型 | 示例 |
|---|---|
| Concept card | RAG 解决什么, 不解决什么? |
| Tradeoff card | RAG vs fine-tuning vs long context? |
| Risk card | Agent excessive agency 如何控制? |
| Artifact card | Requirements-to-Eval Matrix 必须有哪些列? |
| Case card | AML Copilot 为什么不能自动 filing? |
| Interview card | 用 30 秒解释 LLM-as-Judge 的限制 |
9. Monthly Review Template
每月底复制以下表格:
## Monthly AI Asset Review - YYYY-MM
### 1. New assets
- File:
- Capability:
- Evidence status:
### 2. Best learning insight
- Insight:
- Why it matters:
- How it changes my product/BA/architecture judgment:
### 3. Strongest artifact
- Artifact:
- Claim it proves:
- Metric/eval/control:
- Interview story:
### 4. Weakest gap
- Gap:
- Why it matters:
- Smallest next artifact:
### 5. Old asset reused
- Old file:
- New AI reuse:
- Cross-link added:
### 6. Next month focus
- Case:
- Role:
- Artifacts:
- Source anchors to refresh:
10. Quarterly Capability Calibration
每季度给自己打分, 只看证据:
| Capability | Score 1-5 | Evidence | Next upgrade |
|---|---|---|---|
| AI Foundations | |||
| Financial Retail Domain | |||
| BA Problem Framing | |||
| PM Product Strategy | |||
| AI Architecture | |||
| EvalOps | |||
| Governance / Risk | |||
| Adoption / Change | |||
| Vendor / Build-Buy | |||
| Interview Storytelling |
评分标准:
- 1: 只知道概念。
- 2: 能复述。
- 3: 能用于一个 case。
- 4: 能产出 artifact 并解释 tradeoff。
- 5: 能跨多个 case 泛化, 并支持面试深挖。
11. Knowledge Debt Register
随着资料增长, 要管理知识债。
| Debt type | 例子 | 处理方式 |
|---|---|---|
| Unread asset | 旧计划里很多未学内容 | 保留, 放入 review queue |
| Outdated source | 工具版本、模型能力、法规时间线变化 | 标 Needs Review, 补新链接 |
| Weak evidence | 只有笔记, 没有 artifact | 补图、eval、ADR 或 memo |
| Shallow story | 能讲概念, 不能讲经历 | 绑定 case 和 artifact |
| No metric | 只有方案, 没有指标 | 补 baseline、threshold、ROI |
| No control | 只有 AI 功能, 没有风险控制 | 补 control pack 和 runbook |
| No adoption | 只交付系统, 没有用户采用证据 | 补 adoption dashboard |
每月最多处理 3 个知识债, 不要贪多。
12. Portfolio Conversion Rules
一份学习资产变成作品集证据, 必须满足 6 条:
- 有明确业务问题。
- 有用户/干系人。
- 有流程或架构图。
- 有 eval 或质量标准。
- 有风险控制。
- 有业务价值或 adoption 指标。
不满足时, 仍然保留为学习笔记, 不强行包装。
Conversion Example
| Raw asset | Missing | Upgrade |
|---|---|---|
| Transformer 论文解读 | 缺业务案例 | 加 AML/RAG/cost/latency 映射 |
| Web3 治理论文 | 缺 AI 连接 | 加 AI governance decision rights 对照 |
| AML Copilot PRD | 缺作品集叙事 | 加 30 秒 pitch, diagram, eval result |
| Eval 代码 | 缺 PM 解释 | 加 release gate memo |
13. Interview Story Bank Structure
建议把未来所有故事按这个结构整理:
Story Title:
Role Lens:
Business Problem:
Baseline:
My Decision:
Architecture:
Eval:
Control:
Business Value:
Adoption:
Tradeoff:
What I would improve next:
Evidence Links:
每个故事都要有三版:
- 30 秒: 用于开场。
- 2 分钟: 用于主回答。
- 10 分钟: 用于系统设计/深挖。
14. Highest-Leverage Next 20 Artifacts
优先级按“能转作品集 + 能提升面试 + 能连接多个学习资产”排序。
| Priority | Artifact | Source files |
|---|---|---|
| 1 | AML Copilot full portfolio pack | docs/abpa/capstone-aml/, AIPA AML assets |
| 2 | AI Foundations 8-paper concept map | docs/ai-foundations/ |
| 3 | Enterprise RAG ADR for financial policy assistant | RAG note, diagram playbook |
| 4 | Payments Exception Agent sequence + tool risk matrix | case portfolio, diagram playbook |
| 5 | LLM-as-Judge eval suite for customer service answers | eval note, ABPA template 04 |
| 6 | Vendor scorecard for AML/KYC copilot | vendor playbook |
| 7 | AI governance release gate framework | governance plan, NIST/EU/ISO anchors |
| 8 | Lending assistant high-risk control pack | case portfolio, ABPA template 05 |
| 9 | Customer service RAG adoption dashboard | case portfolio, template 10 |
| 10 | Build-vs-buy decision memo | AIPA longform, vendor playbook |
| 11 | Role gap map for AI Solutions Architect | role competency matrix |
| 12 | 3-case architecture diagram gallery | diagram playbook |
| 13 | AI interview proof pack | interview bank, portfolio evidence map |
| 14 | Web3-to-AI reuse note pack | old Web3 notes |
| 15 | BIAN-based banking capability map | BIAN anchor, financial cases |
| 16 | AI cost/latency model worksheet | inference optimization note |
| 17 | LoRA/RAG adaptation decision matrix | LoRA note |
| 18 | CoT explanation policy for high-risk AI | CoT note |
| 19 | AI incident response runbook | governance plan |
| 20 | 12-month portfolio README | this file, master portfolio |
15. Operating Rule: Learn Less Randomly, Convert More Deliberately
未来继续新增资料时, 每个新增文件都尽量标注:
- belongs to which capability domain。
- supports which role。
- can become which artifact。
- should be reviewed at what interval。
- links to which case。
简单规则:
If it cannot become an artifact, it is reading.
If it can support a decision, it is an asset.
If it can be shown and defended in interview, it is evidence.
长期目标不是把所有东西学完, 而是让每一轮学习都更接近:
- 能发现真问题。
- 能设计可评估 AI 系统。
- 能控制风险。
- 能推动采用。
- 能证明业务价值。
- 能把复杂能力讲清楚。