week 50/7 answered0/6 days · 0/30 blocks

RAG evaluation, privacy, and training-data security

Interview focus this week: Single LLM call vs. workflow vs. agent, bounding agent loops, MCP architecture & tool poisoning, and training-time vs. retrieval-time poisoning.

Stored on this device only. Nothing you type is uploaded: no account, no server database, no analytics, no third-party script. It lives in this browser's localStorage for this browser.

Measured, not remembered

Days complete: 0/6
Blocks ticked: 0/30
Evidence days: 0/6
Evidence fields: 0/81
Gaps opened: 0
Gaps closed: 0
Days started: 0
Deliverables: 2

D2 D4 S3 S4 S5 M2 M5

Sunday review · week 5

0/7 answered

01What actually works end-to-end right now without manual intervention?

02What did I measure this week rather than assume?

03What broke during my security labs, and what root cause did it reveal?

04Which security boundary (D1–D12) is now enforced in code and covered by automated tests?

05Which system-design or security interview question can I now answer using my own code as proof?

06What is the single highest-priority deliverable for next week?

07What scope creep or low-priority distraction must I explicitly defer?

This week's days

← week 4 week 6 →