Doktor Whatson
@DoktorWhatson · ~420k subscribers · Long-form essay, variable cadence, 10–30 min
View the channelI wanted to make more explainer videos and lose fewer hours in PDFs. That's exactly what happens now.
↳ The problem
Each video on the channel starts from a synthesis of literature: academic papers, regulatory and educational reports, peer newsletters, and personal notes in Obsidian. The average — measured by the creator himself — was 10 hours of research for every 25-minute video. On top of that came an extra hour of book-keeping to correctly track inline citations. The result: two videos a month instead of the four the audience kept asking for. The backlog of ideas never emptied.
↳ What we built
We built a custom RAG (Retrieval-Augmented Generation) system that automatically ingests the sources — academic saved searches via API, a folder of topical PDFs, the inbox where curated newsletters land, and the Obsidian vault via markdown export. Given a topic, the system produces a structured brief: executive summary, numbered key findings, inline citations with exact page references in the source PDFs, and a "gaps" section flagging where additional manual research is needed. Hosted on a Hetzner VPS in Germany (GDPR-friendly), no data leaving the EU.
↳ Tech stack
- Python
- LangChain
- Qdrant
- OpenAI GPT-4
- Hetzner Cloud (DE)
↳ The numbers
Research per video
Before
~10 h
After
~2 h
Videos published per month
Before
2
After
4
Inline citations tracked
Before
manual
After
automated
Monthly infrastructure cost
Before
—
After
€18 VPS
↳ What they say Doktor Whatson
I never thought of it as an "AI replacement". I thought of it as a patient intern who prepares the brief while I think about how to explain the thing. The brief lands, I correct half of it, and the script is essentially written. It's a different way of working.