Daimon, the Bayesian data science agent from PyMC Labs

Daimon Hackathon · PyMC Labs · #AIWeekNY · Sat, Oct 10, 2026 · 307 W 38th St, NYC

Daimon Hackathon in New York

Use Daimon to analyse a company's data, publish a data story, or extend Daimon with a skill or MCP. Every team gets $25 in Daimon credit.

RSVP on Gomry ↗Join Decision AI Discord (required) ↗
  • 3
    Tracks
  • 2–6
    Per Team
  • $1,900
    Course Seat Per Winner

 Registrations

Presented by

  • PyMC Labs
Chapter I

The Brief

The three tracks and who does what on a team of two to six.

  • Objective
  • Tracks
  • Roles
Challenge

Build with Daimon!

Analyse a fictional company's data warehouse, publish a data story from public sources, or extend Daimon with a skill or MCP. Every team gets $25 in Daimon credit.

Tracks

Pick your track

Each track has its own deliverable. What your team makes for it is up to you. Open a track for the full brief.

Track 01

Data Analysis

You get a company's data: 18 months of orders, costs, returns, chat and meeting transcripts. Find what's really going on, beyond Daimon's first answer, and forecast next quarter's revenue.

For analysts and modelers.

Track 02

A Data Blog Post

Pick a question people argue about in marketing, finance, healthcare, or operations, answer it with public data, and publish a short blog post with Daimon as a Marimo notebook. No code needed.

For writers and researchers.

Track 03

Extend Daimon

Teach Daimon a job it can't do yet, and show the before and after. Build a skill for Daimon, optionally with an MCP server.

For tool builders and workflow designers.

Not sure which? Track 01 investigates a company's data we give you. Track 02 uses public data you find yourself. Track 03 adds a new capability to Daimon.

Track 01 · Data Analysis

The board meets Monday. The pack is wrong.

Meridian Supply Co. sells dive gear, oyster sauce and car oil across Europe. Management thinks it's having a great year. You get 18 months of its data and one question: tell the board what is really going on, and what to do next quarter.

Daimon does the analysis in minutes. It will get some things wrong, and catching that is the work.

The goal

Find what's really going on in the company's data, beyond Daimon's first answer, and forecast next quarter's revenue.

Download meridian_data.zip

What you get

One zip of a realistic company that, like any real one, isn't clean: orders, customers, products and costs, returns, ad spend, the finance report the board sees, a warehouse export, the company's chat and support emails, a board memo and seven meeting transcripts. Start with DATA_DICTIONARY.md.

How it works

  1. Download the zip above, or from your team's channel.
  2. Mention @Daimon in your channel, attach the zip, and ask what's going on.
  3. Split up: one question per person, each in their own thread.
  4. Check what Daimon says, and catch what doesn't add up.
  5. Ask Daimon to turn your findings into a short presentation.

What you hand in, by 2:00 PM

Posted in #hackathon-team-submissions: a 5-slide PDF or a short video, plus a link to your notebook or code. Cover:

  1. What's really going on, in three sentences.
  2. What you found, with numbers and one chart.
  3. What Daimon got wrong at first, and how you caught it.
  4. What the board should do next quarter.
  5. Your Q3 revenue forecast: a number, with a low and a high.

The forecast

The judges know what actually happened in Q3 and compare it with your range.

Bonus: a skill, a short set of instructions that teaches Daimon what to check on a company like this.

Example presentation

Here is a finished 5-slide presentation Daimon made from a single message, with made-up figures. A video covers the same points as your track's slide list above.

Judging

Five criteria, weighed together, no points: novelty; a concrete use case; technical depth; communication; reproducibility. Teams are 2 to 6 people. /billing in Discord shows your team's budget.

Track 02 · A Data Blog Post

A story the data actually supports.

Pick a question people argue about, answer it with public data, and publish a short blog post with Daimon: a headline, one chart and a few paragraphs.

You don't need to code. Daimon fetches the data, writes the code and builds the post as a Marimo notebook it publishes for you. You choose the question and check the answer.

The goal

A short post a reader understands in five minutes, with a number that holds up.

See an example blog post

Pick a question

IndustryThe question
Marketing / retailFive years after the pandemic pushed US retail online, is the e-commerce share still rising, flat, or giving some back?
FinanceWhen the Fed moves, how long until mortgage rates follow, and did the latest moves reach housing starts at all?
HealthcareHow fast did Medicare spending on GLP-1 drugs grow, and how much of it is more patients vs more cost per patient?
Operations / logisticsEighteen months into Manhattan's congestion pricing, what did it do to taxi trips into the zone?

Four suggested questions, each with its public data ready to use:

Or bring your own

Bring your own question in one of these industries. Use downloadable public data; scraping websites is unreliable.

How it works

  1. Mention @Daimon in your team's channel with your question and the sources from the suggested datasets below.
  2. Read its first draft, then check it: is the number right, and does a second source agree?
  3. Ask Daimon to make it a short blog post as a Marimo notebook, and publish it.

What you hand in, by 2:00 PM

Posted in #hackathon-team-submissions:

  1. The link to your published notebook.
  2. A 4-slide PDF or a short video: the question; the answer with your chart; how sure you are and what you checked; what Daimon got wrong at first. Include a screenshot of your notebook.

Suggested public datasets

All are free downloads, no login. Give Daimon the link and your question. Tell Daimon to send a User-Agent that names you and a contact, for example team-x/1.0 (+https://github.com/teamx): FRED and other government sites refuse requests without one.

Example blog post

A published blog post, made by Daimon from one message: one chart of the 30-year mortgage rate and a sentence, as a Marimo notebook.

Example presentation

Here is a finished 5-slide presentation Daimon made from a single message, with made-up figures. A video covers the same points as your track's slide list above.

Judging

Five criteria, weighed together, no points: novelty; a concrete use case; technical depth; communication; reproducibility. Teams are 2 to 6 people. /billing in Discord shows your team's budget.

Track 03 · Extend Daimon

Teach Daimon a job it can't do yet.

Every new Daimon thread starts from zero. Give it a capability for a job it does badly today, and show the difference.

The goal

Pick a concrete job, extend Daimon so it does it well, and show Daimon before and after.

Download the template

What you build

A skill: a short file of written instructions that tells Daimon how to do a job: the steps, what to check, what good output looks like.

Optionally, an MCP server: tools Daimon can call, so the heavy, repeatable part is done the same way every time. Use one that's already online (like DeepWiki, which answers questions about GitHub projects) or run your own.

A good skill on its own can win: you're judged on the idea and the difference it makes, not on the code.

Examples: explain a GitHub project to a newcomer (a skill plus DeepWiki); a monthly check on a company's data; a helper for public economic data; collecting and summarising information from the web.

Adding them: run /agent-setup in your team's channel to add your skill or a server to your team's agent. No code needed unless you build your own server.

What you hand in, by 2:00 PM

Posted in #hackathon-team-submissions: a 5-slide PDF or a short video, plus a link to your notebook or code. Cover:

  1. The job, in one sentence.
  2. Before: Daimon doing the job without your extension (a screenshot).
  3. What you built.
  4. After: Daimon doing the job with it (a screenshot).
  5. A link to your skill or code.

Optional: your own server in one command

Download the template (a starter skill and MCP server), unzip it, install uv, then in the folder run:

uv run serve.py

Attach it to your agent

It starts your server, gives it a public address, and prints the message to paste in your team's channel:

@Daimon attach the MCP server at https://<your-address>/mcp to our agent as <team>-tools

No code needed

Adding a skill, or a server that's already online, needs no code.

Example presentation

Here is a finished 5-slide presentation Daimon made from a single message, with made-up figures. A video covers the same points as your track's slide list above.

Judging

Five criteria, weighed together, no points: novelty; a concrete use case; technical depth; communication; reproducibility. Teams are 2 to 6 people. /billing in Discord shows your team's budget.

Roles

Who does what

Driver

Runs Daimon and owns the repo or notebook.

Analyst

Decides what to ask Daimon and checks the numbers that come back.

Writer

Writes the post, project documentation, and the submission write-up.

Reviewer

Reviews the work before it goes in, the way a judge would.

Teams are two to six people. Roles are a suggestion, not a requirement. One person running Daimon with the others deciding what to ask it works well.

Chapter II

The Build

The six steps, the tool, and the walkthroughs.

  • Steps
  • Stack
  • How-to
Six Steps

Idea to demo in six steps

  1. 00Pick

    Choose a track and form a team.

    Every team must join the Decision AI Discord. Find teammates there, then post your team name, members, and track in #hackathon-team-registration.

  2. 01Set Up

    Choose how to run Daimon.

    Use it on the Decision AI server with no setup, add it to your own Discord server and ask in #hackathon-support for a $25 coupon, or self-host the open-source version with your own API keys. How-To has the steps for each.

  3. 02Frame

    Write down what you are building.

    One paragraph: the question you are answering, the post you are writing, or the feature you are adding. This becomes the description in your submission.

  4. 03Build

    Build it with Daimon.

    Run the analysis, write the code, or draft the post in the Daimon thread. Check the output before you use it.

  5. 04Finish

    Turn it into a short presentation.

    A short PDF or video covering what your track's brief asks for, plus the notebook or code behind it. Daimon can make the PDF for you.

  6. 05Submit

    Submit in Discord by 2:00 PM.

    Post your PDF or video and the link to your notebook or code in #hackathon-team-submissions on the Decision AI Discord.

Work in one thread — Daimon keeps the context of the conversation, so the whole team can see and build on what it has done.

Stack

Your toolkit

Daimon runs in Discord. Use it on the Decision AI server, add it to your own server, or self-host it; How-To has the steps for each.

DAIMON

Your fifth teammate

The data science agent from PyMC Labs. It writes and runs code, fits Bayesian models with PyMC, and posts charts and runnable notebooks in the thread. Free to add, and each team gets $25 in Daimon credit.

QR code for https://daimon.decision.ai

PYMC OPEN-SOURCE STACK

Optional tools

Before you start building, check Decision Hub for agent skills that already do what you need.

How-To

Get set up before you arrive

Three ways to run Daimon

Pick the one that suits your team.

  1. Option 01
    Easiest · no setup

    Use our server

    1. Join the Decision AI Discord server.
    2. Post your team registration in #hackathon-team-registration: your team name, members, and track.
    3. That's it. We'll use Daimon to set up your team from there, and you can start building with Daimon right away.
    Join Decision AI Discord ↗
  2. Option 02
    One-click install

    Your own Discord server

    1. Join the Decision AI Discord server.
    2. Post your team registration in #hackathon-team-registration: your team name, members, and track.
    3. Create your own Discord server for your team.
    4. Add Daimon to your server with the one-click install at daimon.decision.ai.
    5. Ask in #hackathon-support for your team's $25 Daimon coupon.
    daimon.decision.ai ↗
  3. Option 03
    Bring your own keys

    Full self-host

    1. Join the Decision AI Discord server.
    2. Post your team registration in #hackathon-team-registration: your team name, members, and track.
    3. Get the open-source code from github.com/pymc-labs/daimon and follow its README to deploy Daimon yourself.
    4. Bring your own API keys. The $25 coupon doesn't cover a self-hosted copy.
    pymc-labs/daimon ↗

Questions about any of these? Ask in #hackathon-support.

Extending Daimon
Track 03's brief covers skills, MCP servers and the one-command server template. Run /agent-setup in your team's channel to add a skill or server to your team's agent.
Using your team credit
Each team gets $25 in Daimon credit. Start with focused questions and check outputs before running more work. Ask the organizers how to check usage or add credit; the open-source version can also use your own API keys.
Getting help
For setup, credit, or challenge questions, ask in #hackathon-support on the Decision AI Discord. Event support contacts are still being confirmed.
Watch Daimon workThe product site shows a full analysis in a Discord thread, from the question to the notebook.daimon.decision.ai →
Chapter III

The Day

October 10 in New York. Doors at 11, out by 4. The timetable, the rules, and where to register.

  • Schedule
  • Rules
  • Register
Schedule

How the day runs

When
Saturday, October 10, 2026 · 11:00 AM – 4:00 PM EDT
Where
307 West 38th Street, 17th Floor, New York, NY

  1. Doors open · Registration · Team formation · Coffee

  2. Kick-off: Daimon and the three tracks

  3. Building session

  4. Lunch is served (keep building)

  5. Submissions close · Presentations due in #hackathon-team-submissions

  6. Judges review the presentations and walk the tables

  7. Three finalists present for five minutes each · Winners

  8. Event concludes

Rules

The ground rules

  1. R01

    You must use Daimon. The managed version and a self-hosted open-source install both count.

  2. R02

    Teams of 2–6. No solo submissions. Every team must join the Decision AI Discord. One track per team, named in your submission.

  3. R03

    Track 01 analyses the provided dataset and presents what is going on, what you found, what Daimon got wrong at first, what the board should do, and a Q3 revenue forecast with a range.

  4. R04

    Track 02 publishes a data-backed blog post as a Marimo notebook on marketing, finance, healthcare, or operations, and presents what it found. Public data sources only. No scraping.

  5. R05

    Track 03 extends Daimon with a skill, optionally with an MCP server (one already online, or your own from the template), and presents Daimon doing the job before and after.

  6. R06

    Every finding must include the code and data that produced it.

  7. R07

    One submission per team, posted in #hackathon-team-submissions on the Decision AI Discord: a short presentation as a PDF or a video, following your track's brief, plus a link to the notebook or code behind it.

  8. R08

    Submit in #hackathon-team-submissions by 2:00 PM.

Register

Get your team in

  1. 01

    RSVP

    Reserve your seat on Gomry.

    QR code for https://www.gomry.com/event/Hackathon-Build-Multiplayer-Data-Science-Agents-AIWeekNY-9NMINKq8Yz9aAcEUgkJF
  2. 02

    JOIN DECISION AI DISCORD

    Required for every team. Join the Decision AI Discord for team formation, questions, and hackathon announcements.

    QR code for https://discord.gg/umy5TWRTQ
  3. 03

    REGISTER YOUR TEAM

    Post your team name, members, and track in #hackathon-team-registration on the Decision AI Discord so prizes reach the right team.

Chapter IV

The Prize

How judging works, the prizes, and the FAQ.

  • Judging
  • Prizes
  • FAQ
Judging

How we pick winners

Judges look for work that is novel1, answers a concrete use case2, shows real technical depth3, communicates4 clearly, and could be reproduced5 by someone else.

Judges weigh all five together as they walk the science fair and read each Discord submission and Daimon thread.

  1. 1Novelty

    How original is the question, approach, or result? What does the project add beyond an obvious first attempt?

  2. 2Concrete use case

    Does the project solve a clear problem for a real audience, and does the result help that audience act?

  3. 3Technical depth

    How much care and skill went into the method, implementation, and validation?

  4. 4Communication

    Can someone outside the team understand the problem, the work, and the result?

  5. 5Reproducibility

    Can someone else inspect the evidence and follow the steps behind the result?

Prizes

What's up for grabs

  1. First place
    Course seat
    Everyone on the winning team gets a seat in a PyMC Labs data science or analytics course, worth $1,900 each.
  2. Second place
    $50
    Every member of the second-place team gets a $50 Amazon gift card.

Every team gets: $25 in Daimon credit.

Featured projects: The most interesting projects will be featured on the PyMC Labs website, in newsletters, and on LinkedIn.

FAQ

Still have questions?

Eligibility

I'm not technical — can I compete?

Yes. Tracks 01 and 02 don't need you to write code: Daimon writes and runs it, and you decide what to ask and check the results. Track 03 focuses on skills, optionally with an MCP server, that extend Daimon; it does not require a contribution to the PyMC codebase.

Tooling

Do I need my own API key?

No. Every team gets $25 in Daimon credit. You can also run the open-source version of Daimon on your own API keys.

Data

Can I scrape websites for Track 02?

No. Scrapers get IP-blocked and the data is unreliable. Use downloadable datasets, SEC filings, government data portals, and public APIs.

Submission

What must I submit?

One short presentation per team, as a PDF or a video, plus a link to your notebook or code. Post both in #hackathon-team-submissions by 2:00 PM. Each track's brief lists what to cover, and Daimon can make the PDF for you.

Judging

How are finalists picked?

Judges weigh novelty, concrete use case, technical depth, communication, and reproducibility across the science fair, your Discord submission, and your Daimon thread. Three finalists then demo live for five minutes each.

Still have a question?

Ask in #hackathon-support on the Decision AI Discord. RSVP on Gomry to reserve your seat.

Join Decision AI Discord ↗RSVP on Gomry ↗