Bootcamp / Session 3 / 3.1
Session 3.1 · John Damask, Amroja LLC
Building a personal knowledge base with AI.
More than we can remember. Information, and the life-science literature most of all, grows faster than anyone can search, find, understand, and remember it. John’s answer is a knowledge base of your own, kept by an AI agent, and these slides set up the hands-on that follows in 3.2.
The talk in one sentence
Let an AI agent be the custodian of what you read, so you can spend your attention on the questions.
Atopic dermatitis alone gains more than 3,000 PubMed papers a year, about eight a day, and the evidence also arrives as datasets, code, posters, talks, and your own unpublished data. John uses the same structure for a family trip, a class he teaches, and his consulting business.
What the talk covers
- Two goals. Build a knowledge base with an AI custodian, and research with an AI partner.
- Why today’s tools fall short. Downloads pile up and get forgotten. Indexes, from library catalogs to Google, find the container, while the answer sits inside the paper. And your notes live in a different app from the files they describe.
- Where AI helps. Agents widen the attention bottleneck: they search, read, and organize, so discovery and organization feed one place.
- How it works. A wiki of plain text pages on your own computer, one per concept, linked and cited, kept next to the untouched sources. Notes, insights, and generated views live together. New work flows through a loop of search, retrieve, ingest, and distill, driven by a queue of research questions.
- Which tool. Web chat cannot see your files and desktop apps work from inside a sandbox, so terminal agents such as Claude Code and Codex CLI are the best fit. Keep patient data out.
What you build in 3.2
- A
raw/folder of untouched sources.The evidence - A
wiki/of linked, cited pages the agent writes from them.The knowledge AGENTS.md, the rules every future session follows.The custodian’s job- A research task queue the agents work through.The questions
- A 3-D map of how the ideas connect.The view
- A
create-kbskill to start a new knowledge base on any topic.Reuse
Caveats John flags
- Distilling loses detail. A wiki page is a lossy summary, which is why the originals stay in
raw/. - Check its work. Verify what the agent writes, and use the citations it keeps.
- It adds to your sources; it does not replace them.
- The best models cost more. Run research in small batches.
- Shared wikis need review. Before a team edits one knowledge base, decide who approves changes.
About the speaker
John Damask is the owner of Amroja LLC, with more than 25 years of technology leadership in the life sciences. He builds open source and commercial software and helps life sciences and finance organizations speed up their AI adoption. Previously he was VP of Digital Science at Flagship Pioneering and Global Head of Cloud at Novartis. More at johndamask.com.
After the session
Two minutes of feedback, so the next session is pitched better. It also records your attendance, so fill it in whether you were in the room, on Zoom, or catching up on the recording.