# LLM Knowledge Bases: a practical guide — Ben Holmes, Warp

## Executive summary

The talk outlines a comprehensive pipeline for transforming disorganized raw notes into structured, navigable LLM knowledge bases. The process begins with maximizing raw data capture via voice dictation (suggested over typing) and progresses through automated enrichment steps using AI agents. Key stages include enriching notes by adding timestamps, fixed-list tags, web research backlinks, generating wikis of People/Concepts/Sources, and finally visualizing the entire network structure using graph views. The entire system is designed to run on an automated schedule in a cloud sandbox environment.

## Key takeaways

- Raw Data Capture Priority: The most critical step for building a knowledge base is generating high volume of raw material. Voice dictation (approx. 200 words per minute) is presented as the fastest method for capturing thoughts, even if they are 'scrappy' or rambling.
- The Enrichment Pipeline: Notes can be enriched using an agent skill (e.g., 'enrich note') to add structure: timestamps, tags from a fixed reference list (to prevent agents from inventing new categories), web-researched sources, and backlinks found via key term search.
- Automated Knowledge Generation: The system can generate a wiki structure (People, Concepts, Organizations) from raw notes. This process is highly automated and can be scheduled to run in the background (e.g., daily), ensuring the knowledge base remains fresh.
- Automation and Scheduling: To maintain the knowledge base, the entire flow must run on a schedule using cloud sandboxes (e.g., Oz.dev). The process involves syncing markdown files down, running the agent skill, and syncing the updated notes back up.

## Technical details

- Data Capture: Voice dictation tools like Handy (open-source, local model) and Voice Ink are recommended for capturing raw thoughts quickly without requiring subscriptions.
- Note Structure & Linking: The 'enrich note' skill is used to process notes by adding timestamps, generating tags from a fixed list (to constrain LLM creativity), performing web research on sources, and finding related backlinks.
- Automation Workflow: The recommended automated flow uses the Obsidian headless CLI to sync markdown files into a cloud sandbox (e.g., powered by Docker). The agent then runs the enrichment skill across all notes and syncs the modified content back, allowing for background updates.
- Visualization: Graph views can be generated using HTML and Tailwind by instructing an agent to map connections between nodes. This provides a 'bird's eye view' of the knowledge base, helping identify gaps in thinking.

## Practical implications

- Implement a structured, automated pipeline (CI-like flow) to process unstructured data sources (meeting transcripts, research notes) into a centralized knowledge graph.
- Utilize LLM agents not just for content generation, but for metadata extraction and structural enhancement (tagging, linking, timestamping).
- Leverage cloud automation platforms (e.g., Oz.dev) to run data enrichment tasks on scheduled or event-triggered cadences, minimizing manual intervention.
- Design the input capture phase to prioritize volume and speed (voice dictation) over immediate formatting.

## Topics

LLM Knowledge Bases, Knowledge Graph, Automated Pipelines, Voice Dictation, Markdown, Semantic Web, Warp, Hubble (app), Handy, Voice Ink, Obsidian headless CLI, Oz.dev

Source: https://www.youtube.com/watch?v=I3bpdgFJCUY