Initial#149
Open
NeverTheSame wants to merge 42 commits into
Open
Conversation
Author
|
Ok |
…folder and exclude from Git
- Add comprehensive AGENTS.md for AI agent guidance - Enhance README.md with project structure, Python tools setup, and detailed documentation - Add blog posts about AGENTS.md and Nvidia Blackwell - Update .gitignore to reflect posts-generator directory structure
Co-authored-by: kirillkuklin <kirillkuklin@gmail.com>
Co-authored-by: kirillkuklin <kirillkuklin@gmail.com>
Live-site audit found 24 broken internal links on the homepage. Root causes:
1. index.md hard-coded absolute paths like /devops/ and {{ post.url }},
missing the /BeOps baseurl. Replaced with the relative_url filter so
GH Pages serves them correctly.
2. Category index pages /devops/, /k8s/, /sre/, /ai/ did not exist,
causing every post's category breadcrumb to 404. Added _pages/*.md
for each category with permalink set and auto-listed posts.
Verified with a local Jekyll build + Playwright audit: 0 broken links,
0 missing-baseurl links, 0 category integrity issues.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Adds an OSS site audit using Playwright (+ optional lychee) under tests/site-audit/. Detects: - broken internal links - links missing the /BeOps baseurl - posts unreachable from the homepage in >2 clicks - orphan posts - category index pages that fail to list their own posts Runs in GitHub Actions on every push to gh-pages/main and nightly (.github/workflows/site-audit.yml), failing the build on regressions and uploading the report as a workflow artifact. AUTHORING.md + POST_TEMPLATE.md document the link-discipline rules (always use the relative_url filter, never hard-code /devops/) and the post checklist. AGENTS.md updated with the same rules. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…job-interviews category Changes requested: 1. Sidebar (_includes/toc-date.html) now groups posts by category. Each category from _data/categories.yml is rendered as a top-level entry with its posts nested below in date-reverse order. Posts whose category doesn't match a known slug fall into an 'Other' bucket so nothing is silently dropped. Category index pages are no longer double-listed in the utility-pages section (skipped via the new is_category: true front matter). 2. Homepage (index.md) is now structured as 'Browse by Category': each category shows its description and its posts grouped beneath it, instead of a flat 'latest 5 posts' list. 3. 'Back to home' link added at the top of every post and category page via _layouts/post.html and _layouts/home.html (suppressed on the root homepage itself). 4. New category 'job-interviews' added — _data/categories.yml updated, _pages/job-interviews.md created, AGENTS.md + AUTHORING.md + POST_TEMPLATE.md + .github/workflows/site-audit.yml all updated to include the new slug. Future categories only need to be appended to _data/categories.yml (single source of truth for sidebar + homepage) plus a matching _pages/<slug>.md. Verified locally with the site-audit framework: 0 broken links, 0 missing-baseurl, 0 unreachable/orphan/category issues. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- New 'interviews' category for published interview write-ups - First post: first-person recap of a Staff AI Engineer loop covering pipeline, RAG, evaluation, safety/policy, memory, skills/MCP, and observability - Wire the new category into _data/categories.yml, the audit CI env, and AGENTS.md so it shows up in the sidebar and the audit framework Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…adings - Merge the redundant 'interviews' category into existing 'job-interviews' (the two were never meaningfully distinct). - Move the Staff AI Engineer interview post into 'job-interviews'. - Remove duplicate page-title h1s from category pages and homepage — the body.html include already renders one from page.title, so a leading '# Title' in the markdown produced two stacked headings. - Broaden the job-interviews landing copy to cover both real interview write-ups and prep notes. - Update _data/categories.yml, AGENTS.md, and the audit workflow env to drop the 'interviews' slug. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Adds a gh-pages workflow that, on push touching _posts/**, regenerates src/data/posts.ts in NeverTheSame/beops-main-site (newest-first, with category mapping and word-count-based read time) and pushes the change to main. Skips Sat/Sun in UTC. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Restyles jekyll-gitbook with the beops.site palette (parchment / forest green / warm brown), JetBrains Mono + Inter Tight, dashed dividers, flat 2-4px radii, and a dark 'cabin' mode via prefers-color-scheme. Layout (sidebar + topbar) is left untouched; this is colors + typography + component skin only. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Adds a free-tier RAG pipeline so the post generator can cite prior work and stop repeating itself. Stack (zero recurring cost): - Neon serverless Postgres + pgvector (scale-to-zero) - sentence-transformers/all-MiniLM-L6-v2 (384-dim, CPU, MIT) - Hybrid retrieval: pgvector cosine (exact scan) + Postgres ts_rank_cd FTS + Reciprocal Rank Fusion (k=60) — all inside one SQL function Components: - posts-generator/rag/ package: chunker, embedder, db, indexer, retriever, cli - posts-generator/tests/: unit (chunker, embedder, RRF) + integration (indexer idempotency / incremental / orphan-delete, retriever) + regression-blocking eval fixture (5 queries → expected top-K slugs) - .github/workflows/rag-tests.yml: spins ephemeral Neon branch per CI run with 4h expires_at TTL so cancelled jobs can't leak past the 10-branch cap; concurrency group serializes per ref - .github/workflows/rag-index.yml: incremental indexer on push to gh-pages touching _posts/**, guarded by GH Actions concurrency + PG advisory lock Safety rails: - rag_meta table pins embedding_model / chunker_version / schema_version; indexer aborts with clear error on drift - Smoke test asserts BEOPS_RAG_ENABLED unset does NOT pull in sentence-transformers / torch / psycopg from the generator hot path - Generator integration in openai_worker_4o.py is opt-in (BEOPS_RAG_ENABLED=1), default OFF; existing flows unchanged gitignore: un-ignore source code under posts-generator/ so CI sees it; keep configs/ logs/ produced_posts/ and the py-feedparser venv hidden. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
When BEOPS_TEST_DATABASE_URL was set (CI path), the fixture took the 'return dsn' branch and pytest reported 'did not yield a value' because the function had a yield elsewhere — making it a generator that never yielded on that branch. Switch both paths to yield. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
On a fresh DB the meta-version check ran before schema creation and hit 'relation "rag_meta" does not exist'. init_schema is idempotent so it's safe to call first; the version check then runs against a populated rag_meta and aborts cleanly only on real drift. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
New docs/architecture.md with four mermaid diagrams covering: - system overview (author -> GH -> Neon -> generator -> OpenAI) - indexing sequence (chunk, embed, body_hash dedup, orphan delete) - retrieval sequence (embed query -> search_chunks() -> prompt block) - CI quality gate (ephemeral Neon branch, eval fixture regression gate) README links to it from the Features section. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Make the prior-work RAG layer part of the normal drafting flow rather than an opt-in extra: - openai_worker_4o.py: RAG runs whenever NEON_DATABASE_URL (or DATABASE_URL / BEOPS_TEST_DATABASE_URL) is in the environment. Disable with BEOPS_RAG_DISABLED=1 for one-off runs. Auto-skips cleanly (no error) when no DB is configured. - rag/cli.py: new 'context <seed>' subcommand prints the same prompt block the Python worker injects, so any CLI agent (Copilot CLI, Claude Code, etc.) can consult prior coverage before drafting. - AGENTS.md: documents the mandatory pre-drafting step and the one-off opt-out. - smoke test contract strengthened: importing openai_worker_4o never pulls sentence-transformers / torch / psycopg regardless of env vars (heavy imports stay lazy inside the call site). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Lets an external agent ask 'what have I already written about X?' without ever seeing NEON_DATABASE_URL. The secret stays inside Actions; the prompt block returns as a workflow artifact + step summary. Usage: gh workflow run rag-query.yml -f seed='AI Engineering Coach repo' # wait, then: gh run download <run-id> -n rag-context Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
RAG-grounded draft (consulted rag-query workflow before drafting). Cites prior posts: agentsmd-ate-my-readme, staff-ai-engineer-interview, forward-deployed-engineering-ate-customer-success. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.