Append-only. Format: ## [YYYY-MM-DD] action | description. LLM appends here as the last action on every write, per CLAUDE.md.
Rotated 2026-07-28. Full history from vault inception through 2026-07-28 (523 lines) archived to Archive/Log Archive/log-archive-through-2026-07-28.md. The 3 entries below (the vault rebuild that triggered the rotation) are carried forward so this file isn't context-free at the point of rotation.
Interview Prep/ as the new top-level folder for interview-prep + job-switch material (user directive: rebuild vault into 5 life-area folders — Interview Prep, Learning, Finance, Projects, Life — with Interview Prep the highest-priority/accountability folder). Moved via git mv (history preserved): DSA/ → Interview Prep/DSA/; Interview KB/ → Interview Prep/Interview KB/; System Design/ → Interview Prep/System Design/; new Interview Prep/Coach & Tracking/ subfolder holds Private/Project_2026/{Weekly Tracker, Monthly Tracker, Daily Tracker, Job Switch, Projects Done.md, prep-dashboard.html/json, regen_dashboard.py} + Private/Interview Questions for Project or behaviour/ (→ Interview Questions/) + the Coach symlink (→ memory dir) + synthesis/{Coach Hub, Job Switch Hub, Interview Prep Hub}.md; synthesis/DDIA Interview Map.md → Interview Prep/System Design/DDIA Interview Map.md (paired with the DDIA subfolder it indexes). Fixed all literal (non-Obsidian-resolved) path dependencies: .claude/commands/ddia.md, user-level ~/.claude/commands/coach.md, and Interview Prep/Coach & Tracking/regen_dashboard.py (TRACKER_DIR/COACH_HUB/PREP_HUB/JOB_HUB/STAR_NOTE/KB_* constants + output path + docstring). In-vault [[wikilinks]] left untouched — verified they still resolve under Obsidian's suffix/unique-basename matching (spot-checked System Design/Senior SD Mindset, synthesis/Coach Hub, Coach/project_dsa_sd_curriculum style links). index.md and CLAUDE.md still reflect the OLD structure — intentionally deferred to the final rebuild pass (Task 9: rebuild indexes + rotate log). Remaining phases: Learning (Go/Python/SQL/AWS/FrontEnd/AI & ML), Finance, Projects, Life, stale-file audit/archive, then index/log/CLAUDE.md rebuild.Go/, Python/, SQL/, AWS/, FrontEnd/, AI & ML/ + synthesis/{Tech Stack Overview, Concurrency Deep Dive, LLM & AI Stack}.md (general career-upskill notes, split from Interview Prep's curated interview material per user decision — some intentional overlap accepted); added Learning/Tracking/ for Private/Project_2026/Learning/{Books 2026, Courses & Certifications, Python Benchmark Test}.md. Finance/: Private/Project_2026/Finance/* flattened directly into Finance/ (Bank Statements, Monthly Data, scripts, 3 portfolio/review notes, README) — dropped a redundant Project_2026 middle layer after first pass left it non-empty (root-level portfolio files were missed by the directory-level git mv). Projects/: Private/Project_2026/Projects/Personal Portfolio/ → Projects/Personal Portfolio/ (same flattening). Life/: Private/Project_2026/{Experiences, Health, Year in Review, Goals} + Quick Links.md (general lifestyle/travel/planning/logging, cross-cutting docs like 2026 Goals and Quick Links live here since they span all 5 areas). Also folded Private/Resume/ and Private/Work/ (AskTGE war-story source) into Interview Prep/Coach & Tracking/{Resume, Work Reference}/ since both are interview-prep material (resume polish + STAR sources), not general Private docs. synthesis/ folder now empty, removed. Private/Project_2026/Blog/ (always empty) removed; stray .DS_Store files cleaned. Fixed 3 more literal wikilinks pointing at moved basenames (Job Switch Hub resume link, Projects Done war-story link, Past Experience Q&A related-links) plus regen_dashboard.py's AIML_DIR/AIML_EXTRA constants. Patched 3 memory files (project_portfolio_trades_2026, project_current_week, project_interview_kb) with corrected path pointers + reorg notes — historical narrative left untouched, only live pointers fixed. Private/ now holds only Private/Archive/ (10 files) — pending consolidation into a vault-wide Archive/ in the next pass. Next: stale-file audit + archive, then rebuild index.md, rotate log.md, rewrite CLAUDE.md for the new 5-folder architecture.Archive/ folder created, consolidating: Private/Archive/{Deferred/AWS-ML-Cert, Pre-May11-Plan-Summary} (pre-existing archive, relocated) + new Archive/Stale Flagged 2026-07-28/ holding 4 genuinely orphaned items found by a vault-wide inbound-reference scan (0 hits, confirmed before archiving): Pasted image 20260628140641.png (unreferenced screenshot, vault root), Pasted image 20250516234206.png + ...234943.png (unreferenced screenshots, System Design/Images), React Learning.md (2-line redirect stub superseded by the real React.md, only inbound ref was index.md itself), and Private-index.md (a stale duplicate of the Interview KB master index frozen at 97 pages — live version is 103 pages at Interview Prep/Interview KB/index.md — zero inbound refs, clearly a stray save). Also relocated (not archived — actively used, just misplaced): Pasted image 20260628140849.png was sitting in vault root despite being embedded in Redis.md → moved to Interview Prep/System Design/Images/. Removed empty dirs with no content to preserve: Uncategorized/, Private/Project_2026/Blog/, and Private/ itself is now fully dissolved (all content migrated in Phases 1-5) — the 5-folder rebuild has no Private/ folder anymore. Flagged but NOT archived (borderline, judgment calls left to user): Interview Prep/DSA/Notes /Sliding window part 2.png (unreferenced but may be an intentional unembedded scratch image), Excalidraw/Drawing 2026-06-13 23.28.17.excalidraw.md (unreferenced but Excalidraw canvases are often opened directly, not linked), Learning/Tracking/Books 2026.md + Courses & Certifications.md (empty templates, but intentional tracking scaffolds, not clearly "not required").Interview KB/ → Content/ (direct continuation, not a new thing), merged DSA/ in, migrated System Design's non-duplicate remainder in (Docker/K8s, Senior SD Mindset — both had zero KB equivalent; DDIA/, Databases/, My Notes/, Real Interview SD Questions, RAG & LLM System, ML Feature Store). Deleted the 13 confirmed duplicates + the now-empty System Design/ folder. Rebuilt Content/index.md as the one master index (DSA + SD Concepts 25 + HLD 20 + LLD 11 + DDIA/Databases/My Notes 6 + Python 11 + SQL 17 + Go 14 + Frontend 10 ≈ 150 pages). Behavioral & Past Projects/ (Task 16): merged Interview Questions/ + Work Reference/ + Projects Done.md. Job Switch/ (Task 17): kept the existing 30 Company Target List, added an empty Application Tracker.md for when applying starts. Monthly Plan/ (Task 18): archived the old Monthly Tracker/ (6 months pre-built back in March under the pre-replan calendar, never updated, entirely misaligned with reality) to Archive/Superseded Trackers 2026-07-29/. Built fresh Monthly Plan/August 2026.md — completed-summary + new plan + weekly-checklist pointers + pace tracking table + end-of-month retro stub, one month at a time by design (no more pre-building months that go stale). Coach.md (Task 19): compressed Coach Hub + Job Switch Hub + Interview Prep Hub (3 overlapping files, DSA/SD progress duplicated across two of them, never all updated together) into one file — kept current-state pointer, positioning table, non-DSA checklist, interview orientation reference, and the historical scorecard log; cut everything that duplicated Content/ (SQL/FastAPI/Python/API-design/LLD cheat sheets — all live in Content/ now) or Job Switch//Behavioral/ (company list, STAR stories). Old 3 files archived, not deleted. Main Plan.md (Task 20): split the rarely-changing strategy layer (goal, 28.5h/wk budget breakdown, application funnel math, mock cadence, hard rules) out of Coach.md into its own root file, so Coach.md stays pure current-state. Interactive dashboard (Task 21): loaded artifact-capabilities — confirmed only downloads and mcp capabilities exist for this user, no live-data/persistence capability, so true vault-synced interactivity isn't available. Built the most honest version possible instead: a rich interactive Artifact (pace gauges computed from real dates — 105 days to Nov 11, 80 problems remaining, 11.4/wk required vs 11.5/wk historical actual vs 0/wk during the W06-W09 gap — expandable weekly problem checklists, month tabs, SD progress, DSA history sparkline) with localStorage-only checkbox state, clearly labeled in the page and in Dashboard.md as a fast view layer, not a new source of truth — the vault markdown stays authoritative. Published: https://claude.ai/code/artifact/8f287e7a-d0e4-48ff-8e45-73a9ffe7e07e. Archived regen_dashboard.py + its outputs as superseded rather than half-rewiring its parsing logic to the new merged files. Path fixes (Task 22): /coach and /ddia command files, root index.md, CLAUDE.md (directory map, ownership tables, synthesis pages table) all repointed to the new 6-folder structure.num not in perm (O(n) scan) instead of used[] array (O(1)).Interview Prep/Coach & Tracking/ wrapper — Weekly/Monthly/Daily Tracker, Coach Hub, Job Switch Hub, Interview Prep Hub, Projects Done, Resume, Work Reference, Interview Questions, Job Switch, the Coach symlink, and the dashboard script/outputs all now sit directly under Interview Prep/. Flattened DSA/DSA Techniques for Interviews/ and DSA/Data Structures/ directly into DSA/ (kept Notes / and Images/ as asset subfolders, not reading material). Flattened System Design/Concepts/ and System Design/Problem Designs/ directly into System Design/ (kept Databases/, DDIA/, My Notes/, Images/ — legitimately distinct sub-tracks, not taxonomy clutter). Interview KB/ left untouched (103 files, genuinely needs its depth, already has its own index). Fixed every literal path dependency again: /coach, /ddia, regen_dashboard.py, Coach Hub.md's current_week_file pointer, root index.md, CLAUDE.md. (2) Corrected tracked status: DSA/DSA Topics.md's progress table was stale since May (showed W02-W05 as undone despite all PASS STRONG) — fixed to show real state: patterns 01-09 done through W05, Backtracking (10) onward not started. System Design/index.md rewritten with honest status per user: Concepts half-read (unmarked), problem designs done through URL Shortener only (Alex Xu book). (3) Restart: old W06-W10 Weekly Tracker files (created on schedule but zero problems ever logged — the accountability gap from the earlier audit) archived to Archive/Superseded Trackers 2026-07-29/. New Weekly Tracker/Week 06 RESTART (Jul 29–Aug 4) — Backtracking.md created, same problem set, redated to start today. Coach Hub's Current Week Pointer + a new "RESTART" section updated to match. (4) Built Interview Prep/Dashboard.md — single-file daily/weekly task-tracker view (today's checklist, this week's checklist, verified status table, a v1 draft 7-week restart schedule through ~Sep 15, and the Nov-11-timeline tension stated plainly with 4 options, explicitly left for user discussion rather than decided unilaterally). Not committed yet — more refinement expected this session per user ("I will list my pain points").Finance/ moved back under a Private/ wrapper → Private/Finance/ — the one exception to the fully-dissolved Private/, kept specifically for financial data; updated index.md and CLAUDE.md (directory map, ownership tables, exemptions list) accordingly. (2) User said remove (not archive) the 4 borderline items flagged-but-kept in the stale-file audit: Interview Prep/DSA/Notes /Sliding window part 2.png, Excalidraw/Drawing 2026-06-13 23.28.17.excalidraw.md, Learning/Tracking/Books 2026.md, Learning/Tracking/Courses & Certifications.md — deleted via git rm, zero inbound references confirmed first. Learning/Tracking/ now holds only Python Benchmark Test.md.index.md with explicit reading order to every reading-material folder: Interview Prep/DSA, Interview Prep/System Design, Interview Prep/Coach & Tracking (accountability hub, includes a note flagging the W06-W09 unlogged-retro gap found earlier this session), Learning/Go, Learning/Python, Learning/SQL, Learning/AWS, Learning/FrontEnd, Learning/AI & ML, Projects, Life (Finance already had an excellent self-documenting README.md, left as-is rather than duplicated). Root index.md rewritten as a slim top-level map: Meta → 5 folders (each pointing at its own index) → Archive → Raw Sources; page_count 129→156 (includes the new index pages + net vault growth since the last count). Rotated log.md (523 lines → Archive/Log Archive/log-archive-through-2026-07-28.md, fresh file seeded with just today's rebuild entries). CLAUDE.md rewritten next to match the new 5-folder architecture and ownership rules — see the following entry (kept separate since CLAUDE.md changes are self-exempt from the log-on-every-write rule, but this one is significant enough to record).@abstractmethod section to Interview Prep/Content/Backend/Python/Python OOP & Data Model.md (user hit a quiz question on it, gap found — page named abc.ABC/@abstractmethod in the Abstraction pillar bullet but had zero code example or enforcement mechanics). New section covers: full Shape/Circle example, TypeError on instantiating an incomplete subclass, how ABCMeta enforces at instantiation time, duck-typing-vs-ABC tradeoff, and ABC (nominal) vs typing.Protocol (structural) distinction. Added matching interview drill Q4.Interview Prep/Content/Backend/Python/Python Decorators.md (same audit as ABC gap — user asked to check for similar missing quiz-shaped topics). New section covers @a @b @c bottom-up application vs outermost-in execution with a print-tracing example, plus matching drill Q4. Also identified 3 more gaps not yet addressed pending user call: __slots__ (zero mentions vault-wide), descriptors (__get__/__set__, how @property works under the hood — never explained), general metaclasses/type() (only the narrow ABCMeta case exists).Interview Prep/Content/Backend/Python/Python OOP & Data Model.md per user request (prepping for 5yr senior SWE Python/FastAPI/SQL interviews): __slots__ (memory trade-off, PointDict vs PointSlots example, weak-ref/multiple-inheritance caveats), Descriptors (__get__/__set__/__set_name__, Positive validating-attribute example, explicitly ties back to how @property and ORM fields work under the hood), Metaclasses (type as class factory, custom Meta(type) example, ties back to ABCMeta from the earlier ABC section, metaclass-vs-class-decorator guidance). Added drills Q5-7. Audited FastAPI.md, Pydantic.md, Asyncio.md, Identity/Mutability/Memory.md, Scoping & Closures.md, Concurrency.md, Gotchas Cheat Sheet, and the SQL folder (Transactions & ACID, Indexes & Query Performance) — all already at senior depth (isolation levels, deadlocks, EXPLAIN ANALYZE, N+1, TaskGroup, discriminated unions, DI chaining, response_model as an authz boundary) — no further gaps found there.Interview Prep/Content/DSA/09. Heap & Priority Queue.md (gap found — zero mentions vault-wide; surfaced when reviewing a pasted LC 347 solution that only had the heap O(N log K) approach). New subsection under Pattern 1 (Top K Elements): why frequency-bounded ranking enables bucket-indexing, full O(N) LC 347 code, heap-vs-bucket-sort decision rule (bucket sort needs a small bounded-integer ranking key; heap generalizes to any comparable key), and a note that LC 451 (Sort Characters by Frequency) uses the identical trick.Interview Prep/Content/DSA/11. Graphs.md against advanced/senior-level graph coverage (user request: "explain all the patterns, graph is advanced"). Found and fixed real gaps: (1) DFS-based topological sort was missing (only Kahn's BFS version existed) — added with reverse-postorder explanation. (2) Undirected cycle detection was PROMISED in a note ("track parent") but the code was never written — filled in, with explanation of why parent-tracking replaces in_stack for undirected graphs. (3) Added Pattern 8 Bellman-Ford (negative weights, negative-cycle detection, K-limited relaxation variant for LC 787). (4) Added Pattern 9 Floyd-Warshall (all-pairs shortest path). (5) Added Pattern 10 Bipartite check (BFS 2-coloring). (6) Added Pattern 11 Multi-source BFS (LC 994 Rotting Oranges). (7) Added Pattern 12 MST — Kruskal's (reuses existing UnionFind class) + Prim's (structurally mirrors Dijkstra). (8) Added a shortest-path decision table (BFS vs Dijkstra vs Bellman-Ford vs Floyd-Warshall) after the Dijkstra section. (9) Added Dijkstra non-negative-weight caveat explaining why LC 787 (Cheapest Flights Within K Stops) is NOT a Dijkstra problem. (10) Expanded Problems table with 5 new rows (LC 785, 994, 787, 1584, 1334) and a closing note on out-of-scope advanced topics (Tarjan's SCC, articulation points/bridges, A*, Euler paths) offered but not added by default. Also corrected Interview Prep/Content/DSA/DSA Topics.md's Graphs pattern table — it had filed "Cheapest Flights Within K Stops" under Dijkstra, which is wrong (needs Bellman-Ford's K-limited relaxation); added rows for Bellman-Ford, Floyd-Warshall, Bipartite Check, Multi-Source BFS, MST to match the expanded Graphs.md.Interview Prep/Content/DSA/11. Graphs.md Pattern 3 (Topological Sort) and Pattern 4 (Union Find) with conceptual "what it is / why it exists" framing plus fully-traced worked examples (user asked for deeper intuition + examples, not just code, on these two specifically). Topo sort: 5-course prerequisite DAG, full Kahn's algorithm step table (queue/pop/order/indegree updates), plus the cycle-case trace (adding 4->0 empties the queue, order stays short, returns []). Union Find: 6-person friend-group scenario, full union-by-rank + path-compression trace table (parent array + groups after each of 4 edges), plus a worked example of union() returning False on edge (0,2) to show cycle detection falls out of the same mechanism used by Kruskal's MST and Redundant Connection (LC 684).Interview Prep/Job Switch/Remote 40 LPA Crash Course.md + published interactive site (https://claude.ai/artifact/QZR4vxA1GYLARB28LTXyKQ). User pivot: remote roles, 40 LPA floor, easiest loops first, prep Oct 1–31, apply from Nov 1. Web-researched remote-India companies + interview processes (GitLab MR review, Deel take-home + concurrency, Atlassian Karat/code-design, Automattic paid trial, Elastic, Remote.com, Postman, Confluent, Chargebee). Site: 31-day plan, 3-wave company list (20 cos) with rounds/prep/sample Qs, 75 DSA problems, LLD, 12 HLD, AI prep (~38 Q&A + 4 AI SD + coding tasks), take-home/concurrency, behavioral, apply/negotiate. Updated Job Switch/index.md, index.md (page_count 157).Interview Prep/Content/System Design/GenAI/ (12 pages), the vault copy of the published web handbook (https://claude.ai/artifact/DjdRMGgFwKH3W6xCeUHBkh), original content as a free ByteByteGo GenAI SD alternative: index (reading order + 8 priority-ordered free resources), GenAI SD Framework, GenAI Vocabulary, GenAI Capacity Math, 7 problems (ChatGPT-style Assistant, Enterprise RAG Q&A, Smart Compose, Code Assistant, Support Agent with Tools, Text-to-Image Service, LLM Gateway), GenAI SD Question Bank. (2) Learning/AI & ML/Karpathy - Deep Dive into LLMs.md built from the full transcript of youtube 7xTGNNLPyMI (pretraining/tokenization/GPT-2 numbers, SFT + labeler-simulation model, hallucination fixes, params-vs-context, tokens-to-think, RL/DeepSeek-R1, RLHF reward-model gaming, interview hook table, timestamps). Updated: Content/index.md (new §2.6), Learning/AI & ML/index.md (5 pages), synthesis Learning/LLM & AI Stack.md (Related links), root index.md (page_count 157→170).Interview Prep/Content/System Design/GenAI/Vector Quantization.md, a reference page for "what is quantization and why for vectors" plus "explain PQ and IVF-PQ in detail": why vector search is memory-bound, scalar int8 / binary / PQ comparison, PQ train → encode → ADC search with numbers (N=100M, d=1024, m=64), OPQ, IVF-PQ build (coarse k-means, residual encoding) + search (nprobe, per-list ADC tables, merge, rescore) with memory math (~7.5 GB vs 410 GB), tuning knobs (nlist ~4–16√N, nprobe, m, rescore depth), IVF-PQ vs HNSW table, operational gotchas, FAISS snippet, interview Q&A. Linked from GenAI Vocabulary (new row), GenAI index (13 pages), Design Enterprise RAG Q&A, Content/index.md §2.6; root index page_count 170→171.Learning/AI & ML/Karpathy - LLM Intro.excalidraw.md (text elements + 14 embedded screenshots) into Karpathy - Deep Dive into LLMs.md, marked 🖼: full 8-step FineWeb pipeline (Gopher, MinHash, C4 filters) as mermaid; 5k chars → 40k bits → 5k bytes → 1.3k tokens example; DJ-knobs / giant-expression framing + bbycroft nano-gpt (85,584 params); llm.c training log decoded (loss, grad norm, lr, MFU, tok/s); base-model "psychology" bullets; gpt-4o conversation = 50 tokens with special token IDs 200264/200266/200265 (o200k tokenizer); InstructGPT "truthful and harmless > helpful" rule; UltraChat topic-map diversity; RL loop 15→4 correct → train on correct+short; DeepSeek-R1-Zero aha moment (pure RL, no SFT); RLHF cost table 1B vs 5k human scores + ordering-vs-RM-scores example; Ziegler et al. 2019 origin paper; links (FineWeb blog, llm.c #677, Hyperbolic, LM Arena, AI News). (2) Created Learning/AI & ML/LLM Frontier Learning Path.md: 25 verified resources in 7 phases (foundations, reasoning models, agents, LangGraph + GraphRAG, multimodal, JEPA/world models, full course CME295/CS336/Berkeley MOOC) + stay-current list + 3-week schedule. Interpreted "jev model" as JEPA. Updated AI & ML index (6 pages), LLM & AI Stack Related, root index page_count 171→172.Learning/AI & ML/LLM Frontier Learning Path.md: (1) the user corrected "jev model" to Jev by TypeSafe AI, not JEPA. Replaced Phase 6 with a researched Jev / System One models section: launch 2026-09-15, non-autoregressive parallel sampler, typed choice/score/bool outputs with calibrated probabilities, RLCD training, 70–500 ms, pricing, API example, vendor 4-workflow eval vs GPT-5.6 Terra / Claude Opus 5, routing + tool-call guardrail use in agent harnesses, 4 resources (TypeSafe blog, Latent Space / YouTube, LangChain harness post, docs + TypeLLM). JEPA kept as a one-line parked pointer. (2) Added a top "AskTGE-aligned shortlist": 3 certifications first (AWS AIP-C01 GenAI Developer Pro with domain weights, Andrew Ng Agentic AI, HF Agents Course) with a warning that certs are on the Main Plan kill list, then 11 resources mapped to AskTGE features (Bedrock multi-agent, KB tools, prompt caching, parallel tools, summarization, evals), plus the list of what was skipped and why. Added a Jev row to GenAI Vocabulary; updated the AI & ML index summary.