Contents
Hiro is a mighty team: we ship a lot for a team of just over 30 people, and hackathons align closely with our values. Winning a hackathon requires ingenuity, scrappiness, and bold vision. Not so different from our day jobs.
This year’s hackathon took place in January over exactly one week. On the Friday before the hackathon week, individual Hiros pitched their ideas to the company to recruit a team of up to 5 members. Then the next week, hacking began in earnest.
- Mon-Thurs: The teams hacked on their projects
- Friday: The teams presented their projects in a company-wide demo day.
In total, 23 ideas were pitched, and 24 Hiros came together to hack on 12 of those projects. Want to see some of the things we built?
AI Teaches Clarity
Learning a new programming language is hard work. Ideally, you want to look at real-world working examples to learn from, but Clarity doesn’t have a great resource that walks you through existing Clarity code from popular contracts.
Two different teams hacked on LLM projects to address that gap and built apps that take popular on-chain contracts and explain what’s going on in the code in an approachable way. Both projects had lists of popular smart contracts with LLM-generated copy explaining the code.
The first project “Clarity Codex” was built by just one Hiro (!), and as you click through the project’s LLM-generated references, the webpage will direct the reader to the specific lines of code in the contract being talked about, making it easier to learn interactively and click through the contract.
The second team’s project Clarify took a different route. While it didn’t display the code of the contract, it did have an LLM-explainer, but it also showed the contract’s various dependencies. It also enabled you to search for any contract by its on-chain address.
AI Helps You Find Information
Three teams worked on building AI tools that can help you find relevant information in three very different contexts.
The first team “Hiro Rag” built an internal assistant for Hiro resources that can answer employee questions like “how did we fix a recent bug?”, “what are the testnet faucet addresses we use?”, or “who is the most active Hiro on Slack?”
To build the assistant, the team fed all internal Hiro docs, Slack conversations, and more into a private vector database. The chat assistant would then scan that database and provide responses to team queries. In just a week, the team had a fully functioning chatbot, and the link was shared with the entire company for the team to try. Not bad for a week’s work.
The second team worked on smart notifications for the Stacks Explorer. There’s far more on-chain transactions than you can keep up with, and this project introduced a system that can help surface specific transactions that are relevant to individual users.
In this project, every on-chain transaction was categorized through a feature vector and then assigned a confidence score, guessing how relevant that transaction is to a given user. The system then uses reinforcement learning to improve over time—what notifications that the user clicks on vs dismisses are fed back into the notification system.
The last project (a one man team!) built a website that can help you generate Stacks addresses, multisig addresses, convert Stacks addresses to Bitcoin addresses, generate vanity addresses, and more.
Every generated address from this site comes complete with public keys, private keys, seed phrases, and derivation paths.
AI Becomes Hiro’s New Marketing Hire
In this project, 4 Hiros came together to create two “zines”, digital magazines to help market Stacks:
- Hodl Combat: a Mortal Kombat styled guide that compares various synthetic Bitcoin assets and gives each a different “fighter” persona.
- Stacks Field Guide: a survival-themed booklet for new Stacks developers, connecting basic Web3 building principles to camp and wilderness lessons.
In this project, all of the themes, visual assets, and most of the copy were generated by various AI tools, which the team then refined and formatted into a polished product. The team even built an interactive website (with bolt.new) for Hodl Combat where you can pick a fighter with a unique move set and actually fight other assets in turn-based combat.
In the future, you may see these zines appear at various Stacks and Web3 events, so keep your eye out!
On-Chain AI Agent Tells Jokes for STX
One group built an interface in the Hiro Platform that enables users to spin up their own AI agents that could handle a wide variety of tasks, such as calling other contracts, transferring tokens, or even telling jokes in the Hiro #watercooler Slack channel.
To get a joke sent to the company slack channel, all the user has to do is send STX to the agent’s address, along with a word in the memo that will become the subject of the joke. For example, if the memo included “dog”, it told a joke about dogs. And like everyone in Web3, the agent was hustling for its payday:
AI Improves Stacks Analytics
Two projects focused on improving analytics for the Stacks chain and building AI agents that can help parse that on-chain data. Both projects were built by 1-person teams!
In the first project, one Hiro built a data visualization tool for on-chain activity. This tool displayed interactive charts of Stacks activity over time, such as the number of unique sender addresses, block creation, contract deployments, contract calls, and more. An AI assistant helped interpret the charts, highlighting trends and anomalies below each chart.
In the second project, a different Hiro built a Parquet generator and indexer. First, this Hiro extracted blockchain data from a TSV archival file and converted the data into Parquet. Then he launched an AI agent on AWS that could scan that Parquet data and respond to questions like “when was the Stacks genesis block?” or “what was the block hash of the most recent chain reorg?”
AI Builds Better Devtooling
Surprise, surprise, the devs that work at a devtools company work on building better devtools at the internal hackathon. Three different teams worked on devtooling across the stack.
Working our way up from the bottom, one legendary dev (in just a few hours) built “Stackup”, a Stacks node management CLI. Leveraging shell script generated by Chat-GPT, this project makes running Stacks infra as easy as possible. When using this tool, it will automatically install dependencies and download and set up configs for Stacks and Bitcoin nodes. It can even run a signer.
Next up, 3 devs cooked up M.A.N.H.A.T.T.A.N., or Modern Applications of Next-gen Hiro API-based Technologies and Tools for Advanced Networking. That’s a mouthful.
In this project, the team built a new “store and pull” architecture for API services that allows APIs to quickly sync and stream events from arbitrary starting points. Today, syncing a node from genesis is very slow and is a painful process. Event replays are tricky, and deployments are expensive.
This architecture solves that issue and enables multiple nodes to feed data into the “store” part of the service, where events are kept in Postgres. Then any service can connect to this interface and pull both current chaintip data and historical data from a “pull” service.
This service can support multiple APIs syncing at the same time, all from different block heights. We may have more to show you on that front soon…
Last but not least, a team of Hiros worked on building a unit test generator for Clarity contracts. In their project, they built a web page that enables you to paste in a contract code, and then the page generates unit tests for every function in your contract. It even sets up a web container to run the tests in real time to validate that the generated tests are working.
Devs can then copy and download the test file and import it directly into their project for future use.
Conclusion
In just a week, we built all of that, and in the coming months, you may even start to see some of these ideas integrated into our products. Indeed, some of them are already in the works…
If hacking with us sounds fun, check out our careers page. We’re hiring!
