More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    LlamaIndex RAG: Build Retrieval-First LLM Apps

    1

    Build RAG that actually answers from your documents — load, chunk, index, retrieve, and synthesize with citations, tuned at each stage.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    LlamaIndex RAG: Build Retrieval-First LLM Apps

    LlamaIndex RAG: Build Retrieval-First LLM Apps

    Example session with this skill installed

    Our support docs are a mess of PDFs and web pages. Build me a LlamaIndex RAG app that answers refund-policy questions from our own docs, with citations, and tell me how to know it is working.

    • Read your context and instructions
    • Compiled the llamaindex rag

    Load: pointed SimpleDirectoryReader at the docs folder with recursive=True and an explicit extension filter, then added metadata to every document — source filename, URL, date, section, and access level — so retrieval could be filtered by department and date later. Chunk: set 768 tokens with 15% overlap and structure-aware splitting, so headings stay with their sections and no table is split mid-row. The refund policy landed in two clean chunks instead of five fragments. Index: built a vector index as the default and a keyword index alongside it for exact error codes and policy numbers, fused as a hybrid — most real queries mix concepts and identifiers. Persisted the index to disk so deploys stop rebuilding it. Retrieve: top-k 4 with a metadata filter for current policies only, then a cross-encoder rerank over the top 20 down to 5 — the single biggest quality jump after chunking. Synthesize: the query engine answers only from retrieved context and says plainly when the answer is not there; every claim carries its source document title. Asked 'What is our refund policy?' it returned the 30-day window, the exclusions, and two citations. Measurement: ran 20 known questions, hit rate 18/20 in the top-k, and hand-checked faithfulness on 10 answers — all claims supported, and the two misses were chunking problems, fixed by re-chunking rather than prompt archaeology. Conversational follow-ups now condense history into a standalone query first, so 'what about the second one?' retrieves correctly.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Most RAG demos work once and fail in production, because the failure lives upstream of where the symptom shows: bad answers usually mean bad chunking, not a bad model. This skill teaches LlamaIndex as a five-stage pipeline — load, chunk, index, retrieve, synthesize — with deliberate choices at each stage: metadata on every document so retrieval can be filtered, structure-aware chunking at 512–1024 tokens with overlap, vector plus keyword plus hybrid indices chosen by query shape, a tuned top-k funnel with reranking, and an answer contract that says when the answer is not in the corpus. It covers conversational RAG by condensing history into a standalone query before retrieving, and it measures the result with faithfulness checks, retrieval hit rate, and abstention rate instead of vibes.

    It teaches the LlamaIndex pipeline and its tuning decisions, not a hosted search product — you still run the vector store and the model yourself.
    Retrieval quality depends on the corpus: contradictory, stale, or unpermissioned documents produce confident-sounding wrong answers unless metadata filtering and abstention are enforced.
    Reranking costs latency; the skill shows how to measure whether the quality gain justifies it rather than assuming it does.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Almost certainly not. Suspect chunking first, then retrieval, then the index. Re-chunk and re-index is cheap; prompt archaeology is not. Only after the right passages reach the model should you blame the model.

    Vector for 'find me the passage about X', keyword for exact terms, codes, and names, hybrid when queries mix both — which is most real queries. Start hybrid and simplify only if measurement says you can.

    Two parts: prompt the synthesizer to answer only from retrieved context and to say when the answer is not there, and measure the abstention rate on questions you know are unanswerable. If it never abstains, it is guessing.

    New in LLM Ops