Free to read

Retrieve without embeddings

Parse PDFs into a hierarchical document tree. Let a LangGraph agent walk the structure and reason about which sections answer the question. Ship retrieval with zero vector infrastructure and citations your users can actually verify.

Start readingPreview curriculum

Still deciding? Ask first.

Message a mentor about fit, prerequisites, or where to start. Replies come on WhatsApp, usually within a day.

  • Curriculum fit, prerequisites, or where to start
  • Honest answer, no pressure to enroll

Taught by an engineer who has shipped this

  • ISO 27001

    Led the engineering work behind the certification of a regulated EU platform.

  • Series A platform

    Architected the case-management product that became the business a €11.6M round was raised on.

  • 3x faster deploys

    Cut time-to-deploy by migrating to Kubernetes on GCP with deploy-on-merge.

Retrieve without embeddings using hierarchical document trees and LLM reasoning. Zero vector database, zero infrastructure cost. Build a LangGraph agent that navigates PDFs section by section and returns cited answers.

Skip the vector database. Navigate document trees with an LLM agent and get cited answers.

What you'll ship

Real projects, not toy demos.

  • A layout-aware PDF parser that produces a hierarchical TreeNode structure
  • A DocumentTree cache that serializes to JSON and reloads instantly
  • An LLM-generated summary per section so routing stays cheap
  • A LangGraph state machine with analyze, descend, retrieve, and generate nodes
  • Cited answers with page ranges, section titles, and a traversal path

What you'll learn

You finish able to:

  • Explain exactly where vector RAG breaks for structured technical documents
  • Parse PDFs into a hierarchical DocumentTree with layout-aware markdown conversion
  • Cache document trees to disk so repeated questions skip expensive parsing
  • Generate per-node summaries that make routing decisions cheap and accurate
  • Design a LangGraph state machine with conditional routing and path tracking
  • Return grounded answers with page ranges and section titles as citations
  • Measure cost and latency of a vectorless pipeline against a vector baseline

Curriculum

From a raw PDF to a cited answer with zero vector infrastructure.

  1. 01

    Why vectorless

    Understand exactly where vector RAG breaks and why document structure is often the better signal

  2. 02

    PDF to DocumentTree

    Parse PDFs into a hierarchical TreeNode structure with layout-aware markdown and JSON caching

  3. 03

    Section summaries

    Attach a short summary to every node so the agent can route without reading full section text

  4. 04

    LangGraph retrieval agent

    Build a state machine that analyzes nodes, routes conditionally, retrieves content, and generates cited answers

  5. 05

    Production polish

    Visualize the workflow, extend to multiple documents, and instrument cost and latency

Who it's for

Is this for you?

RAG engineers

whose chunked pipelines return near matches but never the right section of a multi-page document

AI backend engineers

who want to ship retrieval without running and paying for a vector database

Applied researchers

building question answering over research papers, legal filings, or technical manuals with real structure

Senior engineers evaluating RAG architectures

who need a principled comparison between vector search and structure-aware retrieval

FAQ

Common questions.

  • Is vectorless RAG always better than vector RAG?

    No. Vectorless shines on structured documents with clear hierarchy (papers, manuals, contracts) and small to medium corpora. Vector RAG still wins for large unstructured corpora where semantic similarity matters more than document structure. You will learn the decision criteria.

  • Do I need a GPU or a vector database?

    Neither. The pipeline runs on CPU with PyMuPDF4LLM for parsing and any OpenAI-compatible LLM for reasoning. The only infrastructure is an API key.

  • How many LLM calls does each query make?

    Typically one call per node visited during traversal plus one final generation. For a well-structured paper, that lands around four to six calls per question. You will instrument the pipeline and see exact numbers.

  • Can this handle multiple documents?

    Yes. The final module covers multi-document trees where the agent chooses which document to descend into before walking its sections.