The Knowledge Wiki

You can paste a PDF into Claude and get a great summary. But next week it's gone. You paste it again. And again for the next fund. And again when a colleague asks about something you already researched six months ago.

The knowledge wiki solves for this. Ingest a document once — fund factsheet, quarterly letter, research paper, your own notes — and it stays queryable forever. You don't search through files manually. You ask questions and get answers drawn from everything you've ingested.

Anything you'd want to reference again. In practice, this means:

Fund docs — factsheets, quarterly letters, DDQs, pitch decks, subscription docs. Research papers and market commentary. Articles you've bookmarked but never properly processed. Your own notes from calls, meetings, and IC discussions.

If you'd paste it into Claude to answer a question, it belongs in the wiki.

The wiki has two layers. Understanding the separation is important — it's what makes querying reliable.

Raw is what was ingested. Wiki is what you know. Keeping them separate means you can always trace an answer back to its source — and update your knowledge pages without touching the originals.

After adding new content, you rebuild the search index. The index is what makes queries fast and accurate — it does more than keyword matching, it understands meaning and context.

knowledge-base/
├── raw/              # immutable source docs — never modify these
│   ├── fund-a-factsheet-q1-2025.pdf
│   ├── market-outlook-2025.pdf
│   └── manager-quarterly-letter-q4.pdf
│
└── wiki/             # your knowledge pages — what you know
    ├── sources/      # one summary page per ingested document
    ├── concepts/     # market concepts, strategy notes, frameworks
    ├── entities/     # one page per fund, manager, or company
    └── index.md      # master catalogue of every page

You don't browse the wiki manually. You ask questions. The search index does hybrid keyword + semantic matching across all wiki pages — it understands meaning, not just words. Claude then synthesises the results into an answer.

Before searching the web for any investment question, I check the wiki first. If it has relevant content, that's my trusted primary source — the web fills in gaps.

"What are the implications from the malaise in direct lending?"

"What does the factsheet say about NAV calculation methodology?"

"Summarise everything we know about this manager across all docs."

"What did we flag as a concern during the last DD on this strategy?"

"Has the fee structure changed since the original factsheet?"

This is the killer use case. Build a due diligence checklist once — as a markdown file with your standard questions, red flags, and thresholds. Ingest it into the wiki.

Now every time you look at a new fund or direct deal, run your checklist against the ingested docs. The AI checks each item, flags gaps, and notes where the docs don't answer your questions. Consistent, cumulative, never forgets what tripped you up last time.

Over time, your checklist improves. You add a question after a fund surprises you. It gets applied to every future investment automatically.

When a new quarterly letter arrives, ingest it. When a manager sends an updated factsheet, ingest it. The wiki accumulates — it doesn't overwrite. Each new document adds to what's already there.

If a manager publishes conflicting information across documents, the wiki captures both versions. You can ask 'Has the fee structure changed since the original factsheet?' and get a direct comparison.

If you're not on the full stack yet, Claude Projects is a good approximation. Upload your fund docs to a Project, write a system prompt explaining what you're doing and how you want it to behave. Claude will search the uploaded files when you ask questions.

The limitation: it's document search against your uploaded files, not a structured knowledge base. You can't query across entities, track changes over time, or run your DD checklist automatically. But it's good enough to start — and it costs nothing beyond your Claude subscription.

Use Projects to validate that the workflow works for you. Once you're pulling value from it daily, that's when the full wiki becomes worth building.

I'm documenting this as I go — the tools, the setup, what worked, what didn't. If you want to follow the build, follow me on LinkedIn. I post updates there as each piece gets built, and I'll prioritise what gets documented next based on what people actually want to see.