How to Build a Research Knowledge Base

The research problem

You read 10 articles about a topic. A week later, you remember the conclusions but not the sources. A month later, you remember vaguely reading something about it. Six months later, you read the same articles again because you can't find your notes.

A research knowledge base solves this by turning raw reading into structured, permanent knowledge. Every article you read gets summarized. Every summary gets linked to related concepts. Every concept builds on previous ones. The result is a personal encyclopedia on your research topic that you can query instantly.

What goes in a research knowledge base

Source summaries. For every paper, article, or report: a 3-5 sentence summary, the key findings, methodology notes, and your personal assessment of quality and relevance.

Concept articles. One article per concept. "Chain-of-Thought Prompting" gets its own page that synthesizes everything you've read about it from multiple sources. This is where compound knowledge lives.

Synthesis pieces. Articles that compare and contrast multiple sources on a theme. "Three Approaches to AI Safety" draws from 8 different source articles and presents a unified analysis.

Open questions. What you don't know yet. What contradictions exist between sources. What needs more research. This is the most valuable section because it directs your future reading.

Literature maps. Visual or textual maps showing how sources relate to each other. Who cites whom. Which findings agree or conflict. Where consensus exists and where the field is divided.

To understand the schema file that governs how your AI builds and queries the wiki, read our explainer on what CLAUDE.md is and how to write one. You can also automate parts of your research pipeline using AI workflow automations like the competitive intelligence monitor.

The CLAUDE.md schema for research

# [Research Topic] Knowledge Base
# Version: 1.0

## Purpose
A structured research knowledge base on [topic],
synthesized from academic papers, industry articles,
and primary research.

## Structure
- /Sources — Raw papers, articles, reports (one file each)
  - /Sources/Papers — Academic and technical papers
  - /Sources/Articles — Blog posts, news, industry analysis
  - /Sources/Data — Datasets, surveys, statistics
- /Wiki — Synthesized knowledge articles
- /Queries — Research questions and synthesized answers

## Categories
- Concepts (one article per core concept)
- Methods (methodologies, frameworks, approaches)
- Findings (key results and evidence)
- People (researchers, organizations, companies)
- Open Questions (gaps, contradictions, areas for more research)

## Article Format
- H1 title
- Summary (must stand alone — someone should get the key insight from the summary)
- Body with H2 sections
- Evidence section (which sources support which claims)
- Counterarguments (what challenges this view)
- Sources section with full citations
- Related articles with [[wikilinks]]

## Rules
1. Never modify source files
2. Every claim must cite a specific source
3. Note disagreements between sources explicitly
4. Flag your own opinions separately from source findings
5. Date every article — research findings have a shelf life
6. Mark confidence levels: [HIGH], [MEDIUM], [LOW] for claims with varying evidence

How researchers query the wiki

"What do my sources agree on about [topic]? Where do they disagree? What evidence supports each position?"

"Summarize everything I know about [concept] from all sources. Include confidence levels and source quality."

"What gaps exist in my research? What questions have I not yet found answers to? What should I read next?"

"Write a literature review section on [theme] drawing from my wiki articles. Include citations."