Pizza, Python, and Prompts: Inside the TypeDB Hackathon
To power into 2026, we turned ‘back to business’ into a celebration, combining Q1 strategy sessions with a hackathon to get the creative juices flowing.
Top priorities: wrap up some threads from last year, flesh out our upcoming plans across the company and stack, and kickstart development with pizza, drinks, and a two days of project-building.
Two hackathon rules: You had to use Claude Code for AI-assisted development and, of course, build with TypeDB.
Why a hackathon
Firstly: they’re loads of fun.
But more importantly, it gives everyone the freedom and permission to take time and explore using the product we spend our day in, day out working on. Sometimes, you get too close and can’t see the issues from up close. And taking that step back already gave us so much real, productive feedback on the product and its UX.
As a side bonus, the hackathon got everyone excited about being back at work and kicked minds into gear: ready to start the year on a high note!
Why AI powered
The engineering world has been shifting with every passing day. The use and power of AI-assisted development is made clearer and clearer. We had some incredible demos and got inspired by our own users and customers – showing us just how much can be built with the right mindset, infrastructure, and frameworks.
That’s where the mandate to use Claude Code came from. This was in part to help overcome internal skepticism, partly to ensure people know what tools are available and going to become part of the standard workflow in the next era of computing.
Our team has the usual mix of opinions on AI tools – some enthusiasm, some doubt. It depends on how deep in the stack we work. For core database engineering, we’re always going to be cautious given how critical it is. But for websites, UX experiments, cloud tooling? There’s probably something there.
CTO demo gets ripped apart
Joshua (our CTO), kicked off by demonstrating his personal Christmas project.
He had Claude Code spend about 12 hours running continuously, upgrading the TypeDB 2.0 C# driver to TypeDB 3.0 protocols. It tracked down and fixed its implementations’ memory leaks, race conditions, upgraded the test framework, and rewrote docs with minimal intervention. It was the perfect setting: a type safe languag, well defined spec with tests, and reference implementations in Java, Python, and Rust already done. The outcome was really impressive!
Then… our engineers ripped the implementation to shreds in their review. And honestly? That really helped calibrate our expectations.
Now, just like the rest of the engineering world, we need to strike a balance: remain good stewards of our codebases, especially those that are safety-critical and must be held to the highest standards. Database systems certainly falls in that category: any sneaky mistake or unconsidered state can cause data loss or corruption.
Right now, AI can get you to 80% completeness and correctness fast, but going from 80% to 100% is still safest with a human involved. Even there though, working with an AI pair programmer does let you move much faster.
What was made
By the end of our two day period, the team made a range of applications. They’re at various different stages of polish, depending on if people decided to refine them further afterwards (several projects did get more work afterwards!). We’ll talk about individual projects more in future blogs since they’re interesting to many of our users.
Here are some of the highlights.
🎧 Unstructed to Structured Audio Knowledge Graph
A working pipeline that transforms raw audio into structured, queryable data in TypeDB. Github repository
🕸️ Prompt-Guided STIX Knowledge Graph Explorer
An LLM-powered backend where users submit natural-language questions to receive answers derived from a STIX knowledge graph. Github repository
🔎 Text to Semantic Search
Automatically build and evolve a structured database for semantic search from a set of text, like your notes! Github repository
🤺 Rusty Foil
Implementing FOIL (First Order Inductive Learner) using TypeQL to learn horn clauses from data. Github repository
🍩 TypeDB Donuts
A fully autonomous donut economy simulation. Watch a central factory supply donuts to retail outlets competing on High Street via real-time exchange. Github repository
📹 Video To TypeDB
Use Claude’s vision capabilities to extract entities and spatial relationships from video scenes and store them in a TypeDB knowledge graph for semantic querying. Github repository
📃 Text2TypeQL
A conversion of Neo4j’s text2cypher dataset, working on presenting a dataset of TypeQL queries for each natural language question. Github repository
🎲 Some random D&D note thing
Storing and displaying the structure of a D&D in TypeDB. Github repository
What we learned
Everyone got practical experience with AI tooling, and we found loads of motivation to set up workloads that can be investigated or implemented in the background.
Agentic programming requires a mindset shift. You stop writing code and start designing testing loops, compile-time checks, and self-validating systems that allow LLM agents to run autonomously. Flip side: if you don’t set up the right infrastructure, you spend your whole day just giving Claude feedback every two minutes.
On the TypeDB front, we found approximately 10 small feature requests and fixes to be made. These include tweaks across docs and error messages, studio features, and one bug in TypeDB Core (~already squashed). An incredibly productive outcome!
Are we doing it again?
Heck yes! For those two days, everyone on the team got to be real TypeDB power users. Even though engineering time is pricey, the team had fun and got comfy with a new tool, and we walked out with valuable feedback for the business,
All that to say – we’ll definitely run another one this year. And we’d love it if you joined us in building fun tools and projects next time!
Sign up for our newsletter or Discord to keep an eye out for announcements, and we’ll give you a heads up when it gets scheduled.