Katze

Crowdsourced data annotation powered by Cryptocurrency

Comment 3

💡 Inspiration

Machine learning is growing at a rapid pace - and we all know it. ML is the next major technological and computational advancement, and big tech companies have created computational resources, making training AI extremely easy. However, discrete or complex cases are difficult to test, meaning annotated data is required.

Annotated data provides the building blocks for ML innovation, improving upon AI metaprogramming, and is integral to training in general. However, getting annotated data is generally difficult, since you may need to manually hire contractors, and try wrangling OSS to get the job done, making training discrete data inaccessible.

Katze initially started as an app for training ML models but we soon realized that this app is needed in the world for an everyday use case, and hence the development of Katze began

🔎 What it does

Katze (the German word for "cat", pronounced /katzuh/) is a PaaS (Platform as a Service) for crowdsourcing data annotation powered by Dogecoin. Customers are able to create Campaigns, which are requests for labeled data.

There are two main components of Katze Campaigns:

Tasks: Tasks are an easy way for campaign managers to quickly get things done with a nice reward for users. Tasks are assigned in the campaign dashboard where managers can select a reward value that will be delivered in crypto to whoever completes the task. Tasks can be of any difficulty and any reward.

Monetization: Katze helps people earn money by spending only 5 seconds of their time. When a user starts a campaign, they are prompted to invest any cryptocurrency of their choice, like Dogecoin, Bitcoin, or Ethereum, into the campaign. We have a blockchain implemented as an Express server that interacts with the Coinbase API to send and receive transactions through the user’s wallet or coin base account. When a user opts in to join the campaign and provide data, a part of that initial investment is sent to them through their Coinbase wallet.

Full Feature List

  • Clean and predictable UI, the low learning curve
  • Ability to start your own campaign with many limits and conditions
  • Power to invest and earn cryptocurrency through campaigns
  • Get data for anything you need on a HUGE scale

🚧 How We Built it

  • Frontend: Lucia.js, TailwindCSS, UrQL, GraphQL, Parcel, TypeScript, Sass
  • Backend: built with GraphQL, PostgreSQL, TypeGraphQL, Prisma, Apollo Server, Express
  • Infrastructure: Netlify, Railway, GitHub
  • Machine Learning: Natural-node for logistic regression classification and basic named entity recognition.
  • Blockchain: All common cryptocurrency transactions with Coinbase API
  • Branding, design and UI: Figma, Adobe Illustrator, Adobe Photoshop, Affinity Designer, Sketch
  • Pitch video: Adobe After Effects, Adobe Premiere Pro, and Adobe Photoshop

🛑 Challenges we ran into

  • Merge conflicts with Git and lag with peer programming software
  • Configuring Parcel and TailwindCSS to work with our bundling process
  • Learning GraphQL and applying it in a meaningful manner
  • Determining the scope of the product
  • Issues with TypeGraphQL and Prisma
  • Implementing the blockchain was a huge pain

✅ Accomplishments that we're proud of

  • Biggest app we’ve ever made
  • Implementing a whole blockchain transaction functionality
  • Helping people by allowing one party to get data and the other party to earn money.

📖 What we learned

  • This product can be impactful to people who want to innovate but need to find a common problem among the people to solve,
  • Working on something that you really enjoy and you know is useful is super exhilarating
  • Learned how to use GraphQL, Railway, Lucia.js, TailwindCSS, and how to cry without noise

💭 What's next for Katze

  • We want to work with AI companies to deploy this in real life
  • We already talked to senior software engineers they say they like our app

Made with 😢 + ❤️ by Tejas Agarwal, Aiden Bai, Melinda Chang, William Lane

Built With

+ 4 more
Share this project:

Updates