Enterprise DNA
Directories / Use Cases / Build a RAG Pipeline

Use case

Build a RAG Pipeline

Wire an LLM to your internal documents so it retrieves the right chunks, generates grounded answers, and does not hallucinate facts your team already wrote down.

The failure mode is this: your team has hundreds of internal documents, PDFs, runbooks, and knowledge-base articles, and every time someone asks the agent a factual question it either makes something up or says it does not know. RAG fixes that by splitting the problem in two: retrieve the right chunks from your corpus first, then generate against only those chunks. The genuinely hard parts are not the retrieval itself but the ingestion pipeline that keeps the index fresh when docs change, the chunking strategy that does not shred meaning across boundaries, and the evaluation loop that catches when retrieval quality quietly degrades. Engineers who have shipped this describe it less as one integration and more as operating a second database alongside the LLM. This stack covers all three of those hard parts.

The stack

Each pick is a real entry on the index. Click any one for the full detail page.

  1. 1
    O OSS RAG framework

    LlamaIndex

    by LlamaIndex

    Why this: LlamaIndex owns the ingestion, chunking, and query-engine layer. It handles PDF parsing, node post-processors, and the retriever-reranker pairing that most teams hand-roll badly. The data connector ecosystem means you connect Notion, Confluence, or a local folder without writing a bespoke loader.

    Full entry
  2. 2
    O OSS Vector store

    Chroma

    by Community

    Why this: Chroma runs locally with zero infrastructure overhead for a first cut, then scales to a hosted instance when the corpus grows. The Python client is the same in both environments, so the ingestion code does not change when you graduate from laptop to prod.

    Full entry
  3. 3
    M MCP Agent retrieval surface

    chroma-core/chroma-mcp

    by Various

    Why this: Once the index is populated, this MCP server exposes Chroma collection queries as first-class tools the agent can call. The agent can search the corpus mid-turn without you writing a custom tool wrapper around the Chroma client.

    Full entry
  4. 4
    A Agents Driver

    Claude Code

    by Anthropic

    Why this: Claude Code runs the ingestion script on a cron, re-indexes changed files, and handles the question-answer loop. The hook system lets you enforce a hard rule: the agent must cite a retrieved chunk ID before answering a factual question, structurally preventing confabulated answers.

    Full entry
  5. 5
    O OSS Evaluation

    Ragas

    by Community

    Why this: Ragas measures retrieval precision, answer faithfulness, and context relevance without manual labeling. You run it on a sampled question set after each ingestion batch. When faithfulness drops you know the chunking or retrieval changed, not just that users are complaining.

    Full entry
  6. 6
    O OSS Relational + vector store (alternative)

    pgvector

    by Community

    Why this: For teams already on Postgres or Supabase, pgvector stores embeddings in the same database as the source metadata. One fewer service to operate, and you can join document chunks against relational filters in a single query, which matters when your corpus has access controls.

    Full entry
Why we picked this stack

Get this running with Enterprise DNA.

Enterprise DNA closes the operational gaps that make RAG pipelines slip quietly. The ingestion job runs as a scheduled project in OPM so you know when it last succeeded and how many documents were indexed. Secrets for the embedding provider and vector store live in Infisical, never in the script. When the Ragas evaluation run flags a faithfulness drop, the alert routes to the inbox rather than a dashboard nobody checks. The whole pipeline is one OPM project with jobs, not a collection of shell scripts in someone's home directory.

Run this stack on Enterprise DNA
Free Blueprint

Get the Stack Blueprint

A printable architecture card with every tool, role, and rationale on one page.

Enter your email. We send one useful update per week. Unsubscribe any time.

Alternative stacks

Different angles on the same outcome.

Alternative

Build a research agent

If the corpus is the open web rather than internal docs, a research agent with Brave Search MCP and a synthesis skill is the right shape instead.

See the alternative Alternative

Build a customer support agent

Customer support over a knowledge base is a RAG pipeline with a customer-facing chat layer in front. Same retrieval core, different surface and escalation rules.

See the alternative Alternative

Build a personal email assistant

When the retrieval target is your own email and project history rather than a shared document corpus, the personal assistant stack handles the inbox surface and draft loop.

See the alternative

Other use cases

More curated stacks from the index.

Use case

Build a customer support agent

A working customer-support agent that triages tickets, answers from your docs, and escalates with full context.

See the stack Use case

Build a research agent

An agent that watches sources, synthesises findings, and ships you a briefing on the days something matters.

See the stack Use case

Build a sales outreach agent

An outreach agent that drafts personal-feeling email, qualifies replies on the phone, and updates the CRM without anyone copy-pasting notes.

See the stack