---
title: "Build RAG that holds up in production"
description: "Composable ingestion, multimodal retrieval, and agentic search — engineered for the production questions basic RAG gets wrong."
canonical: https://ocho.bot/retrieval
last_updated: 2026-08-25
---

# Build RAG that holds up in production

> Composable ingestion, multimodal retrieval, and agentic search — engineered for the production questions basic RAG gets wrong.

Composable ingestion, multimodal retrieval, and agentic search — engineered for the questions basic RAG gets wrong.

## State-of-the-art RAG still gets 37% of questions wrong.

On [Meta’s CRAG benchmark](https://arxiv.org/abs/2406.04744), frontier LLMs without retrieval score below 34%. Straightforward RAG gets you to 44%. Industry-best RAG lands at 63% — with a 17% hallucination rate on the rest. The 37-point gap between that and correct is the entire engineering challenge of production retrieval.

- **34%** — No RAG
- **44%** — Basic RAG
- **63%** — Industry RAG

4,409 questions · five domains · [CRAG Benchmark, NeurIPS 2024](https://arxiv.org/abs/2406.04744)

## Retrieval as infrastructure.

Not a feature you bolt on — the retrieval infrastructure under Ocho, designed for the hard questions under production load.

### Composable primitives

Parsers, chunkers, enrichers, embedders, extractors, and rerankers live in a registry. Each dataset gets a strategy composed from the catalogue — not a fixed pipeline wearing a RAG costume.

### AI strategist per dataset

An agent analyses a representative sample, reviews every primitive with benchmark scores, and proposes a full ingestion recipe with written rationale. Runs once per dataset at $0.50–$2.00 — not per query.

### Eval-gated blue / green deploys

Every strategy change is a hypothesis. New indexes build alongside the live one, run against a golden query set, and only promote on pass. Rollback is a one-line database update.

### Multi-modal: text, graph, visual

Text embeddings for prose, knowledge-graph extraction for relational questions, ColPali-family visual retrieval for figure-heavy corpora. All fused via reciprocal-rank at query time.

## Eight layers. Every call.

Intent, routing, memory, retrieval, tool selection, reasoning, policy, learning. Every call runs the stack; every layer answers to agents.

1. Intent
2. Prompt router
3. Memory
4. Super RAG
5. Tool router
6. Reasoning
7. Policy & guardrails
8. Continuous learning

## Three places this matters.

### Messy PDFs, parsed correctly

Dense tables, merged cells, scanned pages, multi-column layouts. Confidence-gated parsing detects when the primary parser failed and escalates to a vision-language model — so retrieval isn’t bottlenecked on mangled input. Evidence: [kapa.ai — 100+ production RAG teams](https://www.kapa.ai/blog/rag-best-practices)

### Figure-heavy corpora, seen

Charts, diagrams, S-N curves carry information OCR can’t extract. ColPali-family multi-vector embeddings index full page rasters, fused with text retrieval at query time. The Strategy Agent enables this path only when figure density warrants it. Evidence: [ViDoRe Leaderboard](https://huggingface.co/spaces/vidore/vidore-leaderboard)

### Multi-hop questions, answered

Comparisons and cross-document synthesis break single-shot RAG. Knowledge-graph extraction builds typed relations at ingestion time, so traversal joins evidence across docs with provenance back to the source chunk. Evidence: [Microsoft GraphRAG](https://arxiv.org/abs/2404.16130)

## These aren’t marginal improvements.

- **49%** fewer retrieval failures — from contextual enrichment alone — [Anthropic](https://www.anthropic.com/news/contextual-retrieval)
- **67%** with contextual + BM25 + reranking — compounding ingestion-side improvements — [Anthropic](https://www.anthropic.com/news/contextual-retrieval)
- **20–30%** boost from hybrid search — over vector-only via reciprocal-rank fusion — [Weaviate](https://weaviate.io/blog/hybrid-search-fusion-algorithms)
- **3.4×** comprehensiveness on entity-rich corpora — graph-based over vanilla RAG — [Microsoft Research](https://arxiv.org/abs/2404.16130)

## Bring us your hardest retrieval problem.

We’ll map your corpus, score where your current stack breaks, and walk through what a production-grade pipeline looks like against your actual data — whether we work together or not. 30 min · free · no deck, no pitch.

[Book a 30-min Fit Call](https://ocho.bot/contact)

---

Ocho — AI knowledge orchestration · [Home](https://ocho.bot/) · [Docs](https://ocho.bot/docs) · [Blog](https://ocho.bot/blog) · [About](https://ocho.bot/about) · [Developers](https://ocho.bot/developers) · [Contact](https://ocho.bot/contact) · [llms.txt](https://ocho.bot/llms.txt)