Apr 29, 2020
Clever Sheets wins global Slack Hackathon using Aito
In the beginning of March, Slack organized a virtual global hackathon in which teams were challenged to develop the most modern and engaging app experience for their users .
Using the Aito predictive database, Intwixt won the competition in the most challenging category: Best Net New Directory App with their Clever Sheets app. Criteria for this category were to build a fully functionally Slack application during the week of the hackathon.
We spoke to Sabin, who built the app together with Luke Birdeau. Sabin and Luke are co-founders of Intwixt.
Congratulations on winning the hackathon with the Clever Sheets app!
Thank you, winning the Slack hackathon has been amazing news for us. We have received a lot of feedback from different industries and customers. Simultaneously, this feels like the worst time to win a hackathon — there is so much going on with the worldwide pandemic. It hasn’t stopped us from already thinking on how to further develop Clever Sheets though.
Can you tell me a bit about the hackathon and category Best Net New Directory App in which you won?
This was a worldwide virtual hackathon. Upon subscription, we got access to a closed Slack workspace through which we received all necessary info. Everything was very well organized, very detailed and we got amazing guidance from Slack’s product and engineering teams. There were 5 different categories for submission. We chose for the possibly most challenging one, as the Best Net New Directory App meant that the team was required to build a fully functional Slack application during the time of the Hackathon. There weren’t any requirements on a specific language to use, except that the teams needed to use 1 of the 4 features of the Slack App toolkit: App Home, Modals, Granular Permissions, and Block Kit.
Each App was rated on multiple criterias. Slack published them in advance so the process was very transparent.
You and Luke decided to go develop Clever Sheets — an app that can help predict missing values in Google Sheets. How did you decide on that particular idea?
We looked for an idea that is simple enough that everyone would understand. For a way to put AI/ML-technology in the hands of ‘regular’ people and make it more widely available. Clever Sheets combines 3 different technologies that are all about eliminating frictions in the workflow for the user.
- Slack: amazing technology for communication
- Spreadsheets (Google Sheets): most common interface used by professionals
- Aito: a predictive database for developers, eliminating feature engineering
We felt that these 3 technologies combined could represent a recipe for success.
We are very excited — Clever Sheets is easy to use for anybody, with any Google Sheet you have. You upload it to Slack and can start making predictions.
What is the main target group for Clever Sheets?
Professional people that use Google Sheets on a day-to-day basis. Basically any company that deals with tabular data, or with spreadsheets that contain categorical columns which might have missing values. In the demo video we used the example for a company that needs to create work-shifts for their employees. This is a great example of how to use Clever Sheets.
How does it work technically? Is there a lot of knowledge required from the user?
Not at all! So a Google Sheets document can have multiple sheets in it. Clever Sheets will use the data from one sheet to train one single model. If two sheets are part of the same document, they will not inference each other. Each sheet is one single Aito table.
The onboarding flow contains a couple of short questions to map which columns are categorical and which columns can have empty values. Based on these questions and answers, Clever Sheets figures out what columns can be predicted by Aito’s predictive database. Clever Sheets looks at all the rows that are complete and those are used to train the Aito model with them.
To make things more user-friendly, there is a colour scheme that highlights the rows that were predicted, with a colour that indicates the confidence level. The confidence level is defined by the user during the onboarding flow. It’s easy to adjust confidence levels afterwards, or even to generate a new sheet and put the 2 sheets with different confidence levels side by side.
Infrastructure-wise, the application was built with Intwixt, which is a low-code platform for building AI-enabled Slack applications. We’re very proud that we were able to build the app in one week. It was a very intense week, though :-) Joking aside, one of the key messages here is that it is much easier to build applications on top of Slack than traditional web applications.
We could have not built this application as a standalone web application at the same quality and in the same timeframe. Another key message here is that in one week we were able to build an AI/ML application that deals with dynamic ML models using Aito. Every sheet is a different ML model, one that gets created and trained in real time when the user uploads the sheet. We could have not done this using a non-autoML type technology.
What would be the most common use cases for Clever Sheets?
Basically any spreadsheet containing categorical data. Eg exports from Salesforce on predicting which sales representative would be best to lead certain deals, predicting the risk level of a project given a set of properties used for measuring success, predicting the likelihood for a defect to be closed in time, predicting the likelihood of a success hire given the profile of the candidate, and so many more…
How do you see the app evolving?
We would love to expand this application to make it even smarter, to make it a very powerful tool for consultants. Consultants spend most of their time in spreadsheets; setting up processes, looking for errors, fixing issues. More than just predicting missing data, it would be amazing to build a tool that helps consultants to work with spreadsheets in a more efficient way.
What are the restrictions on what you can predict?
For the moment, Clever Sheets is only able to predict categorical data. If the values are absolute, those cannot be predicted (yet). Making predictions that go beyond categorical data e.g. “what is the best price for this product” is very difficult. Even predicting the price range is very hard.
Anything more you would like to add?
Yes, here is a message to anyone that is looking at building human workflows, especially workflows that involve web based UIs: do yourself a favour and consider building them on top of Slack. You’ll end up building them faster, on top of a modern platform that provides so many benefits. There is a lot of literature that talks about these benefits, so I am not going to talk about them. I will only say that it takes less time to build a sophisticated UI in Slack, especially if you use BlockKit, the app exposes less friction since it is closer to the user and you get to leverage a great distribution channel.
One last thing — go try Clever Sheets out! The app is free, and will remain free for the time being. There are some restrictions on the free use: maximum 5 sheets per user and 1000 rows per sheet. A video about the workflow can be found here.
Originally published at https://aito.ai.
