- Website https://slayerfest.org
- Summary Fully automated AI academic conference on Buffy the Vampire Slayer
- Collaborators Built with Frederic Kettelhoit
Slayerfest is fictional academic conference about Buffy the Vampire Slayer. It runs a full AI academic workflow — complete with grad students, post docs, peer reviewers, and professors — all bots, no humans (except the original chat log they're sourcing from).
It ingests thousands of lines of Buffy discussion from a real conversation, runs it through a bureacratic multi-agent process, and outputs a "conference proceeding" website with papers, reviews, and revision histories.
The multi-agent system is fully vibe coded. It has a lot of redundancy in it. But so does academia.
| Bot | Description |
|---|---|
| Researcher Bot | Generates potential abstracts and paper ideas |
| Grad Bot | Summarize weekly conversations (no thesis) |
| Post Doc Bot | Rate weekly summaries (0–100) against a thesis |
| Research Assistant Bot | Selects the most relevant Buffy episode scripts |
| Professor Bot | Writes the academic paper |
| Peer Reviewer Bots | Review and force revisions until all papers are accepted |
You can also learn more about what each bot does on the technical documentation page of the conference website here.
Each conference run automatically produces papers, reviews, and randomly versioned filenames (e.g. paper_FINAL_v2_copy3.md).
A full conference can take many hours, because there are many 60 second breaks baked in to manage the 20,000 token/min rate limit. Below is a sped up 10 minute version of the bots at work, with all the rests cut out:
You first need to set up some details in your .env file:
ZULIP_API_KEY=your-zulip-api-key
ZULIP_SITE_URL=https://your-zulip-instance.com
ZULIP_EMAIL=your-email@example.com #email you use for zulip
ZULIP_RECIPIENT=other-person@example.com #email of the other person in the 1:1 conversation
CLAUDE_API_KEY=your-claude-api-key
This project uses uv for dependency management. You can sync with:
uv sync
# 1. Extract conversations
uv run main.py extract-private --limit 10000 --weekly-chunks
# 2. Analyze with grad bots
uv run main.py run-grad-bots --weekly-dir "conversations/weekly"
# 3. Generate paper abstracts (5-8 philosophically-driven topics)
uv run main.py researcher-bot
# 4. Run the full conference (generates papers, reviews, revisions until acceptance)
uv run main.py run-conference
There are some key commands for run-conference, which may significantly impact the output:
--notes-folder: Grad notes folder (default: most recent) --min-abstract-rating: Min abstract rating to include (default: 0) --num-reviews: Reviewers per round (default: 3) --max-revision-rounds: Max revision attempts before giving up (default: 10) --min-rating: Min postdoc rating to include notes (default: 30) --max-scripts: Episode scripts to include (default: 5) --verbatim-chat-threshold: Rating for using verbatim transcripts (default: 70)
Once the papers are ready:
uv run python static_site_generator.py
This takes source files and builds the website from it. You may want to point to a specific paper directory with the option --papers-dir (default papers)
There are some other options here that let you run parts of the pipeline, without a full conference. For instance, you can run a "lab" to generate a paper for a human-provided abstract with
# 1. Extract and run grad student steps above
# 2. Generate paper (auto-runs postdoc bot + research assistant + professor bot)
uv run main.py generate-paper --topic "Your paper topic"
The generate-paper command automatically:
- Runs postdoc bot to rate notes for relevance (0-100)
- Runs research assistant to select relevant episodes
- Runs professor bot to write the paper using filtered notes and scripts