05 — Solution Architecture Patterns
Overview: Design patterns for AI systems in enterprise. You'll learn the anatomy of a reference architecture, where to draw integration boundaries, how to model cost and latency, and what can go wrong.
What This Module Tests
- Architecture depth — Can you design a system that's credible at a whiteboard?
- Trade-off reasoning — Can you defend a design choice when questioned?
- Failure mode thinking — What breaks? What's the blast radius? How do you mitigate?
- Pattern recognition — Can you identify patterns across use cases?
Learning Path
Recommended order:
| Step | File | Purpose | Time |
|---|---|---|---|
| 1 | 01-Reference-Architecture-Anatomy | Components: LLM, retrieval, evaluation, orchestration, HITL | 15 min |
| 2 | 02-Integration-Boundary-Doctrine | Where does the AI system end? Where does the customer's system begin? | 15 min |
| 3 | 03-Failure-Modes-Catalog | Hallucination, latency, cost explosion, data leak, compliance violation | 15 min |
| 4 | 04-Cost-and-Latency-Modeling-Primer | Math: token count → cost; latency SLA → model choice | 15 min |
| 5 | 05-Rollout-Patterns | Shadow mode, canary, gradual rollout, HITL gates | 10 min |
Total notes time: ~70 minutes
Practice Exercises
| Drill | Focus | Time | Rubric |
|---|---|---|---|
| 01-Whiteboard-Support-Agent-Assist | Design a customer support agent on the board; walk through the flow | 15 min | Clarity, completeness, realistic constraints |
| 02-Cross-Architecture-Pattern-Recognition | Given 3 use cases (search, agent, classification), identify the patterns | 10 min | Abstraction, pattern clarity |
| 03-Trade-off-Defense | "Why Sonnet and not Haiku?" "Why not just LangChain?" Defend design choices | 10 min | Logic, honesty, customer awareness |
Goal: All 3 at Hire level minimum. Strong Hire means you never say "YOLO" in an architecture.
System Designs
Production-grade architecture documents for all 6 BFSI use cases:
- 01-Customer-Support-Agent-Assist — HLD/LLD for support use case
- 02-Internal-SOP-Policy-Search — RAG system for policy lookup
- 03-Relationship-Manager-Copilot — Context-aware customer relationship system
- 04-Compliance-Document-Review — Regulatory compliance automation
- 05-Developer-Productivity-with-Claude-Code — Internal developer productivity
- 06-Executive-Analytics-Assistant — Analytics & business intelligence
Running BFSI Customer
- 01-Six-Use-Case-Meta-Pattern — BFSI customer has 6 AI use cases. What patterns do they share? Where are the differences?
Knowledge Check
Interview-QA-Bank covers: - Architecture design (technical questions) - Trade-off reasoning (HM/technical questions) - Failure mitigation (values/technical questions)
👉 Skim the Q&A Bank first. Mark your 5 weakest questions.
How to Work This Module
- ✅ Read all 5 Notes in order (~70 minutes)
- ✅ Run the 3 Drills — Get comfortable at the whiteboard
- ✅ Study the 6 System Designs — These are production reference documents
- ✅ Review the case study — Understand the BFSI customer's 6-use-case landscape
- ✅ Skim Interview-QA-Bank — Mark your 5 weakest questions
- ✅ Mark complete when all Drills reach Hire level
Strong-Hire bar: You can sketch a reference architecture and articulate failure modes under skeptical questioning.
What's Next
After this module: - You can design a credible enterprise AI system - You know where to draw integration boundaries - You can model cost and latency confidently - You're prepared for whiteboard design interviews
Next Module: 06-Evaluation-Architecture
Quick Reference
Key Files: - Notes: 01-Reference-Architecture-Anatomy and 4 more - Drills: 01-Whiteboard-Support-Agent-Assist and 2 more - System Designs: 01-Customer-Support-Agent-Assist and 5 more - Case Study: 01-Six-Use-Case-Meta-Pattern - Interview Q&A: Interview-QA-Bank