Developer Highlights: Projects built with Orion
Jul 28, 2023
Bonjour les amis!
Several Gizabrains were at the annual EthCC event last week; hacking, conspiring, presenting and attending talks. Here’s a comprehensive recap of what was cooking in Paris and some exciting updates from our developer community.
We hosted a ZKML track during the LambdaClass Hackathon, which gave rise to some bleeding-edge projects and implementations by hackers, recapped below in addition to a prize-winning project from ETHGlobal.
Our Paris-native lead engineer Raphael Doukhan hosted several Orion workshops and hack sessions all around the city. If you missed the workshops, you can still tune in via this link, and join the Orion developer channel on Discord to attend future meetups.
Our cofounder Fran Algaba presented verifiable ML in StarknetCC and joined leading researchers and builders in the ZKML space on a panel during the OFR Paris | Generative World Architecture event. Recordings are added below.
Let’s break it down.
✨ Orion Projects
Orion hackers were in full swing during our week in Paris. We highlight in this blog the diverse set of exciting projects that emerged. If you’d like to learn more about Orion before delving into the projects, check out the docs here.
Let’s dive in.
1. Drive AI 🛣️
Drive AI is a road-fighter-inspired simulation environment implemented with Dojo, a provable game engine that enables zero-knowledge proofs to be generated attesting to the integrity of computation ⛩️
How it works:
1️⃣ AI learns to drive in a road-fighter-inspired environment implemented with Dojo.
2️⃣ The cars are controlled using a neural network.
3️⃣ Inference is done with Cairo on-chain, bootstrapping Orion.
Watch Cartridge founder Terrance presenting the Drive AI demo, displaying this ambitious experiment where a fully on-chain physics engine is bootstrapped with neural network inference. Check out the StarknetCC presentation below!
2. Delta Z — LambdaClass Hackathon 2nd Prize 🥈
@manujmishra_ and @ac_crypto won the second prize during Lambda ZK Week with Delta Z, the first zero-knowledge differential privacy protocol, built in Cairo using the Numbers API implementation in Orion.
What is Differential Privacy?
Differential Privacy (DP) offers mathematical guarantees that individual data cannot be deduced from query results. Traditionally it relies on a trusted data curator, which is a single point of failure for vulnerability.
Delta Z is a completely trustless DP solution. The team has implemented key algorithms from Differential Privacy literature as a Cairo-based smart contract deployed on StarkNet (Goerli). This allows users to submit queries as normal but receive results with verifiably correct noise that doesn't hinder their analysis.
3. Einstein Summation in Orion — Best Use Award at ETHGlobal
einsum is a function that was popularized by NumPy. Its name likely originates from the Einstein summation convention, from which it borrows its style for expressing tensor contractions. The Einstein summation convention provides a simplified notation for various tensor operations, specifically contractions. einsum will play a foundational role in the Orion library, as it facilitates the abstraction of multiple operators represented in the runtime, allowing us to achieve higher rates of compatibility.
4. Convolution Operator
Another key operator was implemented during the Lambda ZK Hackathon last week by @codingnirvana.
Convint is a long-awaited operator for the Orion library, which will enable certain computer vision tasks to be performed in Cairo.
Orion Community is Growing!
Our lead developer Raphael Doukhan hosted several workshops throughout the week at the LambdaClass ZK Hackathon, ETHGlobal, StarknetCC and Antalpha Labs Hacker House.
If you missed these workshops and meetups, check out the recording from the LambdaClass ZK Week!
Going forward, we will be hosting many more workshops both online and IRL. If you’re a developer looking to build with Orion, follow the steps below to get involved.
Giza Talks & Presentations
If you missed the presentations and panels we took part in, check out these recordings below!
StarknetCC: Fran Algaba on Enabling Provable Machine Learning with Cairo
OFR Paris | Generative World Architecture: Diving into ZKML