[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79SA54RE85X2X2TEDKHZ0FV":3,"skill-same:01M79SA54RE85X2X2TEDKHZ0FV":53},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":32,"commitSha":33,"skillDirectory":34,"forksCount":14,"repositoryHtmlUrl":35,"repositoryHomepageUrl":6,"ownerHtmlUrl":36,"ownerBlogUrl":6,"ownerDescription":37,"updatedAt":38,"install":39,"resources":52},"01M79SA54RE85X2X2TEDKHZ0FV","rag-architect",null,"Architect-level retrieval-augmented generation design — chunking and parsing strategy, embedding and index choice, hybrid search and reranking, citation and grounding contracts, freshness and re-indexing, and the retrieval evals that separate real grounding from plausible text. Use when answers must come from a private corpus.","SriAkhilSJ\u002FPulseAIRepo","SriAkhilSJ","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F206201543?v=4","01M79RWFD1Q09B89SXTTKKJY6Z","Sri Akhil SJ","01M79RWFE26NPG0BZ1FRNX376H",0,"MIT","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fai-ml-architect\u002Fretrieval-architect\u002Frag-architect\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: rag-architect\ndescription: Architect-level retrieval-augmented generation design — chunking and parsing strategy, embedding and index choice, hybrid search and reranking, citation and grounding contracts, freshness and re-indexing, and the retrieval evals that separate real grounding from plausible text. Use when answers must come from a private corpus.\nlicense: MIT\ncompatibility: pgvector, Qdrant, Weaviate, OpenSearch, Pinecone\nmetadata:\n  crusor-domain: ai-ml-architect\n  crusor-path: ai-ml-architect\u002Fretrieval-architect\u002Frag-architect\n  crusor-origin: original\n  crusor-sources: \"arxiv.org\u002Fabs\u002F2005.11401; github.com\u002Fpgvector\u002Fpgvector; docs.llamaindex.ai\"\n  crusor-triggers: \"rag, retrieval, vector database, embedding, chunking, semantic search, reranking, hybrid search, knowledge base, citations, pgvector, hallucination, hallucinating, grounding, make things up, makes things up\"\n  crusor-verified-on: \"2026-10-30\"\n---\n\n# RAG Architect\n\nRetrieval quality is **the ceiling on answer quality**. A perfect model over bad\nchunks produces confident nonsense.\n\n## When to use\n\n- Answers must be grounded in a defined corpus (docs, tickets, contracts, code).\n- The knowledge changes and cannot live in weights.\n- Answers must cite sources for trust or audit.\n\n## When NOT to use\n\n- The corpus fits comfortably in a cached context window and changes rarely —\n  paste it and skip the retrieval system.\n- The question requires reasoning over the whole corpus at once (aggregation) —\n  that is a query engine problem, not a similarity problem.\n\n## Decision 1 — Pipeline stages, each measurable\n\n```\ningest → parse → chunk → embed → index → retrieve → rerank → assemble → generate\n                                    ↑                                   ↑\n                            retrieval evals                     answer evals\n```\n\nEvery stage has an owner and a metric. Most RAG failures are parse and chunk\nfailures wearing a model costume.\n\n## Decision 2 — Parsing and chunking\n\n| Corpus | Approach |\n|---|---|\n| PDFs | layout-aware parser, keep tables as tables |\n| HTML docs | strip nav, keep heading hierarchy |\n| code | function\u002Fclass boundaries, keep signatures |\n| tickets\u002Fchat | one thread, resolved state first |\n| spreadsheets | row-group chunks with headers attached |\n\n- Chunk by semantic boundary, 300–800 tokens, with 10–20% overlap.\n- Carry metadata on every chunk: source, section, timestamp, tenant, permissions.\n- Never embed a chunk you cannot cite back to an exact location.\n\n## Decision 3 — Embedding and index\n\n- One embedding model per index; changing it means a full re-index — plan the\n  migration before you need it.\n- Dimension and distance metric are architectural constants; document them.\n- Vector store choice by operational fit: pgvector if you already run Postgres\n  and the corpus is modest; a dedicated store when scale, filtering, or hybrid\n  search demand it.\n- Metadata filters are part of the query contract: tenant, permission, freshness,\n  document type.\n\n## Decision 4 — Retrieve well\n\n- **Hybrid**: BM25\u002Fkeyword + dense vectors, fused (RRF). Exact terms, IDs and\n  names fail pure-vector search.\n- Retrieve 20–50 candidates, rerank with a cross-encoder to the top 5–8.\n- Query rewriting for conversational input; decompose multi-part questions.\n- No results is a valid outcome: answer \"I don't know\" rather than filling the\n  gap.\n\n## Decision 5 — Grounding contract\n\n- Every claim maps to a chunk id; render citations with source and location.\n- Assembly: put the strongest chunks first, delimit them as data, and instruct\n  the model to refuse when the context does not contain the answer.\n- Permission filtering happens **before** retrieval results enter the prompt —\n  never rely on the model to redact.\n\n## Decision 6 — Freshness and re-indexing\n\n- Ingestion is a pipeline with idempotent upserts keyed by content hash.\n- Deletes propagate: a removed document must disappear from the index, or it will\n  be cited forever.\n- Re-embedding runs as a background migration into a new collection, then a swap.\n- Track index lag; a stale index is a correctness bug with a delay.\n\n## Failure modes at scale\n\n- Chunks split mid-table; the model reads half a price list.\n- Pure vector search missing an exact error code the user pasted.\n- Citations generated from memory of the prompt rather than returned metadata.\n- Tenant leakage through a missing metadata filter.\n- Deleting a source document that keeps appearing in answers.\n\n## Verification receipts\n\n1. Retrieval eval: recall@k and MRR on a labelled question set.\n2. Answer eval: groundedness and citation accuracy, judged against the corpus.\n3. Chunk inspector screenshot for one hard document (tables, headings intact).\n4. Permission test: a query from tenant B never returns tenant A chunks.\n5. Delete test: removed document no longer retrievable, verified by query.\n","# RAG Architect\n\nRetrieval quality is **the ceiling on answer quality**. A perfect model over bad\nchunks produces confident nonsense.\n\n## When to use\n\n- Answers must be grounded in a defined corpus (docs, tickets, contracts, code).\n- The knowledge changes and cannot live in weights.\n- Answers must cite sources for trust or audit.\n\n## When NOT to use\n\n- The corpus fits comfortably in a cached context window and changes rarely —\n  paste it and skip the retrieval system.\n- The question requires reasoning over the whole corpus at once (aggregation) —\n  that is a query engine problem, not a similarity problem.\n\n## Decision 1 — Pipeline stages, each measurable\n\n```\ningest → parse → chunk → embed → index → retrieve → rerank → assemble → generate\n                                    ↑                                   ↑\n                            retrieval evals                     answer evals\n```\n\nEvery stage has an owner and a metric. Most RAG failures are parse and chunk\nfailures wearing a model costume.\n\n## Decision 2 — Parsing and chunking\n\n| Corpus | Approach |\n|---|---|\n| PDFs | layout-aware parser, keep tables as tables |\n| HTML docs | strip nav, keep heading hierarchy |\n| code | function\u002Fclass boundaries, keep signatures |\n| tickets\u002Fchat | one thread, resolved state first |\n| spreadsheets | row-group chunks with headers attached |\n\n- Chunk by semantic boundary, 300–800 tokens, with 10–20% overlap.\n- Carry metadata on every chunk: source, section, timestamp, tenant, permissions.\n- Never embed a chunk you cannot cite back to an exact location.\n\n## Decision 3 — Embedding and index\n\n- One embedding model per index; changing it means a full re-index — plan the\n  migration before you need it.\n- Dimension and distance metric are architectural constants; document them.\n- Vector store choice by operational fit: pgvector if you already run Postgres\n  and the corpus is modest; a dedicated store when scale, filtering, or hybrid\n  search demand it.\n- Metadata filters are part of the query contract: tenant, permission, freshness,\n  document type.\n\n## Decision 4 — Retrieve well\n\n- **Hybrid**: BM25\u002Fkeyword + dense vectors, fused (RRF). Exact terms, IDs and\n  names fail pure-vector search.\n- Retrieve 20–50 candidates, rerank with a cross-encoder to the top 5–8.\n- Query rewriting for conversational input; decompose multi-part questions.\n- No results is a valid outcome: answer \"I don't know\" rather than filling the\n  gap.\n\n## Decision 5 — Grounding contract\n\n- Every claim maps to a chunk id; render citations with source and location.\n- Assembly: put the strongest chunks first, delimit them as data, and instruct\n  the model to refuse when the context does not contain the answer.\n- Permission filtering happens **before** retrieval results enter the prompt —\n  never rely on the model to redact.\n\n## Decision 6 — Freshness and re-indexing\n\n- Ingestion is a pipeline with idempotent upserts keyed by content hash.\n- Deletes propagate: a removed document must disappear from the index, or it will\n  be cited forever.\n- Re-embedding runs as a background migration into a new collection, then a swap.\n- Track index lag; a stale index is a correctness bug with a delay.\n\n## Failure modes at scale\n\n- Chunks split mid-table; the model reads half a price list.\n- Pure vector search missing an exact error code the user pasted.\n- Citations generated from memory of the prompt rather than returned metadata.\n- Tenant leakage through a missing metadata filter.\n- Deleting a source document that keeps appearing in answers.\n\n## Verification receipts\n\n1. Retrieval eval: recall@k and MRR on a labelled question set.\n2. Answer eval: groundedness and citation accuracy, judged against the corpus.\n3. Chunk inspector screenshot for one hard document (tables, headings intact).\n4. Permission test: a query from tenant B never returns tenant A chunks.\n5. Delete test: removed document no longer retrievable, verified by query.\n",{"name":5,"license":15,"metadata":24,"description":7,"compatibility":31},{"crusor-path":25,"crusor-domain":26,"crusor-origin":27,"crusor-sources":28,"crusor-triggers":29,"crusor-verified-on":30},"ai-ml-architect\u002Fretrieval-architect\u002Frag-architect","ai-ml-architect","original","arxiv.org\u002Fabs\u002F2005.11401; github.com\u002Fpgvector\u002Fpgvector; docs.llamaindex.ai","rag, retrieval, vector database, embedding, chunking, semantic search, reranking, hybrid search, knowledge base, citations, pgvector, hallucination, hallucinating, grounding, make things up, makes things up","2026-10-30","pgvector, Qdrant, Weaviate, OpenSearch, Pinecone","main","8fa4e240be65b98472c9b2d3a505ade2aea1a5c6","skills\u002Fai-ml-architect\u002Fretrieval-architect\u002Frag-architect","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ","Studying in CMR institute of Technology\r\n2024 - 2028","2026-11-13 21:02:23",{"source":8,"skillName":5,"commands":40,"prompt":50,"zipAvailable":51},[41,44,47],{"tool":42,"command":43},"npx","npx skills add SriAkhilSJ\u002FPulseAIRepo --skill rag-architect",{"tool":45,"command":46},"bunx","bunx skills add SriAkhilSJ\u002FPulseAIRepo --skill rag-architect",{"tool":48,"command":49},"pnpm","pnpm dlx skills add SriAkhilSJ\u002FPulseAIRepo --skill rag-architect","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fai-ml-architect\u002Fretrieval-architect\u002Frag-architect\u002FSKILL.md",true,[],[54,61,68,75,82,89],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M79TTAJNACCNXCEN3649CMMD","interaction-patterns-architect","Architect-level interaction design in code — choosing established patterns over invented ones, feedback and latency behaviour, destructive-action safeguards, navigation and overlay conventions, and the state model behind every interactive surface. Use when a screen has to behave, not just look.","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fwebsite-architect\u002Fui-ux-implementation-architect\u002Finteraction-patterns-architect\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M79TT2ESKB1TKZBYY9KYJHV9","design-system-ops-architect","Architect-level operation of a design system in code — token pipelines and versioning, component API and variant policy, the catalogue as the contract, adoption metrics, and the governance that stops a system from forking into three. Use when a component library must survive several teams.","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fwebsite-architect\u002Fui-ux-implementation-architect\u002Fdesign-system-ops-architect\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M79TSV0PWGNK3PRX9P4JDR9N","design-handoff-architect","Architect-level design handoff — what must be specified before implementation starts, tokens over raw values, responsive and content-limit contracts, state matrices, and the review loop that catches drift between design and code. Use when a design is about to become code.","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fwebsite-architect\u002Fui-ux-implementation-architect\u002Fdesign-handoff-architect\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M79TSJS9BFXNP696FMJ9GKDA","content-microcopy-architect","Architect-level UI content — where strings live, voice and terminology consistency, error and empty-state copy as part of the interface, translation-ready authoring, and the content model a product needs before it multiplies screens. Use when the words in the interface matter as much as the pixels.","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fwebsite-architect\u002Fui-ux-implementation-architect\u002Fcontent-microcopy-architect\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M79TSCNBST3N1M68V1KPH9HQ","ui-ux-implementation-architect","Architect-level direction for turning a design into working UI — the handoff contract, component ownership versus screen-level layout, interaction and state conventions, design QA gates, and the review habits that keep an implementation faithful without frozen pixels. Use before any specific UI implementation skill.","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fwebsite-architect\u002Fui-ux-implementation-architect\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M79TS4N6A4D2QQ433VXS5D0R","test-strategy-architect","Architect-level test strategy — allocating assertions across unit, integration, contract and end-to-end layers, designing for testability through dependency boundaries, fixture and data strategy, and the flake and coverage policies that keep a suite trustworthy. Use when defining how a codebase is tested.","https:\u002F\u002Fgithub.com\u002FSriAkhilSJ\u002FPulseAIRepo\u002Fblob\u002Fmain\u002Fskills\u002Fwebsite-architect\u002Ftesting-and-deployment-architect\u002Ftest-strategy-architect\u002FSKILL.md",[],[]]