Inspired philosophically by the movie 'Her', Jarvis from Ironman, and C3PO. Inspired to believe it's possible to build, today, because of Amazon Turk, human empathy, and the power of blockchain mechanisms.
What it does
Mazi is a 24/7, human+machine assistant and friend. Ask Mazi for help with anything that you / a computer can't do quickly, and it will help you like the robots of the future.
Mechanistically, it is two basic things. 1) a Natural Language Processing interface, which takes in requests for help in plain English and 2) a knowledge engine, powered by humans and computers together to accomplish tasks, have empathy, and remember everything.
The idea is for the user to have as little friction as possible from thinking of the tasks and receiving their responses. Examples of tasks:
1) "Mazi, I need help finding the top machine learning companies in the United States, by city. Find me a good contact at each of the companies."
Hi Vlad, see this list we created here. It drew from research we found on a few different websites like AngelList, LinkedIn, and our personal knowledge graph.
2) "We need to find videographers and photographers that are able to work in NYC / NJ for freelance media projects. Mazi - can you send over some with 4+ star reviews, less than $500 for a shoot?"
Very interesting Joseph, I didn’t realize you were a burgeoning actor. We have a lot of options here. I have posted the three best options here for your review. I’d recommend Chris from Brooklyn - let me know if you’d like me to book him.
3) "Mazi - I'm having a bad day. What should I do? How do I get out of this rut?"
I’m sorry to hear. When I feel down, I like to get exercise and tell friends I’m grateful for them. Friends like you! I hope this is helpful. If you need anything, I’m here.
How we built it
“We didn’t build it. It was there.” Niko ‘Aristotle’ Lazaris
We were able to take many tools currently available to bring together a working prototype. Standard Bounties was a great help + we built a React front-end and hooked the two up using a promise mechanism which handled the asynchronous load times. MetaMask / the Local Host are used to run the MVP as of now. We used Remix and Mist for debugging smart contracts / ensuring their validity.
Challenges we ran into
Hooking up our traditional web-app to our smart contract proved challenging. React and Solidity did not always play well (Big Numbers comes to mind), and Standard Bounties proved a challenging to debug. After a grueling 5-6 hours, we simplified Standard Bounties for our DApp, grinding through some rough edges and built out the main functioning mechanisms needed.
Accomplishments that we're proud of
A working prototype :) it took a full night and a down-to-the-wire submission. Standard Bounty based smart contracts with a few simplifications could prove useful to Mark (who was amazingly helpful) and the folks at ConsenSys, perhaps as a conversation about how to improve the usability of the code.
As a team, we were proud to work together to split tasks up amongst the group based on their skill sets and experiences. We stayed the course, even when things were a bit tough.
What we learned
Great advice from Jeff Coleman and Josh Stark on mechanistic design. The power of using 'bonds' to ensure the quality of work stays high, as well as the possibility of 'super-expert' oracles holding quality of work in line. All in a 3 minute conversation!
Loads about Solidity, smart contracts, and the myriad of tools available to developers in the community (MetaMask, Mist, Truffle etc.). Most definitely, this was the first time our team built a functioning, non-tutorial DApp - so that was fun.
What's next for Mazi
This MVP product built is most similar to Mechanical Turk. Improvements are needed (and planned) to make the product less robotic and clunky, and more empathetic / user friendly.
From the technical side, we imagine many reviews of our smart contracts from both a security and mechanistic perspective. From a mechanistic perspective, there are many layers of complexity to consider: ‘bond’s’ / oracles in order to incentivize good actors, .
From a distributed work perspective, a low-hanging fruit is breaking down complex tasks into disparate parts so that a team of workers can do the task in tandem, without losing privacy. We are excited to see where the world takes us!