Day 5 · Fri 2026-10-09 · ChatGPT CORE
Production LLM design: RAG freshness, tenant boundary, context budgets, evals, and fallback
Week 11 trains two designs with one operating spine: admission, isolation, durable work, queue/backpressure, streaming/reconnect, derived state, evals, and recovery.
3contiguous 1-D DP slots
25-35mattempt + follow-up window
Staffinvariant, failure, recovery, metrics
Exact Daily Schedule
No drift: the same six blocks run every day.
| Time | Block | Contract |
|---|---|---|
| 08:50-09:00 | Target | Write one Staff-level invariant and one failure window before reading. |
| 09:00-10:05 | Project design | 25-35m production LLM extension prompt, then RAG/evals/tenant-boundary decision pass. |
| 10:05-10:45 | DDIA / source grounding | Map the design to exact DDIA anchors and only direct applied-AI sources. |
| 10:45-10:55 | Spoken close | Record the architecture in English, then say the Chinese invariant aloud. |
| 14:30-16:15 | 1-D DP contiguous block | Three NeetCode slots below; each slot is 25-35 minutes plus follow-up. |
| 20:30-21:15 | Recall / artifact | Run English Senior+/Staff recall and patch the scorecard evidence. |
Canonical Hello Interview Anchors
Direct day-page anchors only. Source titles were verified live before build.
| Canonical anchor | Disposition | Design decision |
|---|---|---|
| Some additional deep dives you might consider | CORE extension | Use only extensions that alter serving, eval, security, or operations. |
| What is Expected at Each Level | CORE bar | Use level rubric to decide Staff depth. |
| Mid-Level | CORE bar | Mid-level must have a coherent serving path. |
| Senior | CORE bar | Senior owns scheduler, streaming, and context trade-offs. |
| Staff | CORE bar | Staff owns tenant boundaries, eval loops, fallback, and operating controls. |
Exact DDIA Mapping
Only Ch5, Ch9, and Ch13 sections are in scope.
| Chapter | Exact section | Interview use |
|---|---|---|
| Ch13 | 维护派生状态 | Vectors, summaries, prompt-cache keys, and leaderboards are derived state. |
| Ch13 | 观察派生数据状态 | Freshness, ACL rejects, stale retrieval, and drift must be observable. |
Applied AI Source Gate
First-party/original sources appear only when they change a serving, eval, security, or operating decision.
| First-party/original source | Concrete decision it changes |
|---|---|
| OpenAI - Retrieval | RAG freshness, chunking, and source attribution are part of product correctness. |
| OpenAI - Evaluation best practices | Golden datasets and graders gate prompt or model changes. |
| OpenAI - Evaluate agent workflows | Trace-level evals catch multi-step regressions that output-only scoring misses. |
| OpenAI - Data controls in the OpenAI platform | Tenant isolation, retention, and zero-data-retention choices constrain feature design. |
| Anthropic - Effective context engineering for AI agents | Context selection is an engineered budget with tools, memory, and state, not unlimited chat history. |
| Anthropic - Demystifying evals for AI agents | Agent evals should inspect process, tool use, and failure modes, not just final answers. |
| Anthropic - How we contain Claude across products | Untrusted code/tool execution requires sandboxing, egress control, and per-product isolation boundaries. |
21-Slot 1-D DP Runway
Today is slot 13-15 of 21, contiguous across the week.
| Block | NeetCode problem | Mode | Invariant | 25-35m follow-up | Complexity |
|---|---|---|---|---|---|
| 14:30-15:05 | Combination Sum IV | Ordered count | dp[target] sums ways from previous targets, and order matters. | Ask how answer changes if order does not matter. | O(target * nums) / O(target) |
| 15:05-15:40 | Coin Change II | Unordered count | Iterate coins outside and amounts forward to avoid permutation duplicates. | Ask contrast with Combination Sum IV. | O(amount * coins) / O(amount) |
| 15:40-16:15 | Target Sum | Subset transform | Convert signs to subset sum when total plus target is even and nonnegative. | Ask for zeros multiplying counts. | O(n * target) / O(target) |
Artifacts And Recall
Use these in the evening block and scorecard.
ChatGPT Chinese deep lectureChatGPT Chinese lecture audioEnglish Senior+/Staff recall scriptEnglish recall PDF
Required spoken close: explain one invariant, one failure window, one recovery path, and one metric in English. Then restate the same design decision in Chinese without adding new components.
Detailed lecture notes, audio, recall scripts, PDFs, Staff Q&A, and mock packs are archived locally and are intentionally not published on this site.