Algovera | A Decentralized Hub for Data Scientists in Web3 | Round 17

OceanDAO Round 17
#1

Project Name

Algovera

Project Category

Build & Integrate

Proposal Earmark

General

Proposal Description

This month, we’re further integrating storage solutions, expanding machine learning asset definitions (e.g. model weights are currently not defined as an asset) and automating aspects of algorithm publishing. The latter means that data scientists will be able to write an algorithm in a notebook (the environment that they are most used to) and run a script to convert the notebook and publish to the Ocean marketplace. We will also develop a proof-of-concept frontend for a catalogue of machine learning assets. This will be built on the Ocean marketplace, while also supporting open source datasets on IPFS. As usual, we’ll run expansive community activities including another grants round and a first decentralized AI community IRL meetup just after DeSci Berlin.

Grant Deliverables

Tools/Libraries/Integrations

  • [ ] Expand PoC of StorageProvider (using <web3.storage>) to work with <estuary.tech> and HuggingFace storage.
  • [ ] Expand Ocean Assets module for different types of machine learning assets (e.g. datasets, algorithms, weights) and their specifications.
  • [ ] Make algorithm publishing easier for data scientists by creating a tool (using Jinja templates) to automatically parse Jupyter notebooks for assets (e.g. datasets, algorithms, weights), convert to C2D format and upload the assets to the Ocean marketplace.
  • [ ] MVP of app frontend for search and discovery of ML datasets on Ocean or IPFS. App will display both open source and private assets.
  • [ ] Demo video for Estuary integration on JupyterLab
  • [ ] Demo video for Lit integration on JupyterLab

Community

  • [ ] Algovera Grants Round 3
  • [ ] Make landing pages for squads
  • [ ] Run a first decentralised AI meetup in Berlin
  • [ ] Set up <Monday.com> for project management
  • [ ] Design data bounties and data jobs board feature for Algovera Website
  • [ ] Set up <Intros.Ai> for Algovera community relationship building.

Project Description

Algovera is a maturing ecosystem of independent AI teams. AI projects in the Algovera ecosystem are reaching the stage where they can apply for larger funding programmes. For example, the Virtual Object Detector Squad is applying for OceanDAO funding, the DeFi Squad has a working app (video demo, code) and the Healthcare Squad produced a first draft at a whitepaper. You can keep up with progress and links of Squads from Round 1 here. In our second round of grants, we funded 4 more AI projects and 1 bounty for design work on the Algovera app. Algovera provides an infrastructure composed of a growing number of Web3 technologies that makes it easier for community members to progress from initial idea to a team with a monetized AI app. In the last month, we had possibly our biggest month of building yet!

Final Product

A Web3 AI marketplace with tools, libraries and integrations that make it easy to build, collaborate and monetize AI apps.

Value Add Criteria

Data scientists are natural data consumers. Creating further tools, libraries and integrations (and making it easier for data scientists to onboard to Web3) will help to increase consume volume of data and algorithm tokens. Funding more data science projects will result in more new projects and assets using Ocean. Completing freelance work will help to develop raw data into algorithms and insights.  

Core Team

Richard Blythman

Hithesh Shaji

Keaton Kirkpatrick

Jakub Smékal

Funding Requested
20000

Minimum Funding Requested
12000

Wallet Address
0x823351c03A99b4820793675760f4A64F5ccA9089

#2

Im not sure that these deliverable qualify as core tech deliverables and wonder if Algovera should make seperate proposals for separate projects that they work on.

The Community Deliverables do not pertain to core tech development

additionally

  • [ ] Make algorithm publishing easier for data scientists by creating a tool (using Jinja templates) to automatically parse Jupyter notebooks for assets (e.g. datasets, algorithms, weights), convert to C2D format and upload the assets to the Ocean marketplace.

this seems like it will be integrated into Algovers’s Jupyter Library and not an ocean library is that correct?

#3

OceanDAO currently requires one proposal per project.

Very open to discussing core tech status. Can you share the reasons that you don’t think the deliverables qualify as core tech? Currently, you have just stated that you don’t think that they qualify.

this seems like it will be integrated into Algovers’s Jupyter Library and not an ocean library is that correct?

No, that’s not correct. That deliverable will not be part of our Jupyter library. Not sure where yet, but it will be compatible with Ocean libraries. Happy to submit a PR on this in future.

#4

Project submitted deliverables:

  • [✓] Expand PoC of StorageProvider (using web3.storage) to work with estuary.tech and HuggingFace storage. We added our existing PoC for web3.storage to ocean.py and and added functionality for estuary.tech (see here). We also added a class for uploading to HuggingFace Hub to improve discoverability of assets (see here). We also added a readme with a demo flow for encrypting assets and uploading to both decentralized storage and HuggingFace.
  • [✓] Expand Ocean Assets module for different types of machine learning assets (e.g. datasets, algorithms, weights) and their specifications. You can see the code where we’ve added stronger distinctions between different assets in our dHub library. This is actually trivial in ocean.py since there is very little distinction between different types of assets. I think adding more distinction might be useful for data scientists. The best approach for this will likely become more clear when we implement more functionality for ML Ops (e.g. Kubeflow pipelines) in future.
  • [✓] Make algorithm publishing easier for data scientists by creating a tool (using Jinja templates) to automatically parse Jupyter notebooks for assets (e.g. datasets, algorithms, weights), convert to C2D format and upload the assets to the Ocean marketplace (code with guide here).
  • [✓] MVP of app frontend for search and discovery of ML datasets on Ocean or IPFS. App will display both open source and private assets.
  • [✓] Demo video for Estuary integration on JupyterLab
  • [✓] Demo video for Lit integration on JupyterLab
  • [✓] Algovera Grants Round 3 (blog post)
  • [✓] Make landing pages for squads (DeFi Squad)
  • [✓] Run a first decentralised AI meetup in Berlin (blog post)
  • [✓] Set up Monday.com for project management
  • [✓] Design data bounties and data jobs board feature for Algovera Website (figma)
  • [✓] Set up Intros.ai for Algovera community relationship building (blog post)
#5

Admin:

Thank you so much for the great work and awesome updates. It’s easy for me to follow and verify that you are addressing all of your deliverables. As a brief feedback, it would be great to see more links on each item so curators/voters can dig deeper. All the best! -Idiom