From small-scale ML models in graduate school to massive-scale systems at Uber, I’ve trained a lot of Machine Learning models. No matter the model, it’s magical when months of work condense down to a single
predict function call.
Then comes the hard part.
Your sales counterpart gets excited about a lead scoring model that you are working on. Naturally, they want to use it. You quickly realize that a
predict function inside a Jupyter notebook isn’t quite as magical to the sales team. Maybe an API endpoint? You import some libraries, run some commands, and navigate the maze that is the AWS console. Your model is now available as an API endpoint via AWS Sagemaker and API Gateway.
You go back to the sales team and show off this new API. They still aren’t impressed. They spend all day inside Salesforce, not a terminal! You now go back to the drawing board and try to figure out a way to connect this model to Salesforce. You need to spin up a complex system that monitors for new sales leads, scores each lead with your model, and sends them back to the sales team.
That project that you thought was almost done? It now has another 2 months of painful integration plumbing work.
We built Booklet.ai so you can spend a day - rather than months - integrating a model into your company’s data flow. We believe that once you deploy a model, you should be able to start sharing and sending those results immediately. These are the principles behind Booklet.ai:
It is too complex to create a web app for your ML model using Flask+React+Docker, etc.
Web Developers spend years mastering Flask, React, and Docker. You have days to create a web demo for your non-data scientist teammates. You shouldn’t have to create and deploy a web app from scratch for this! Sadly, that’s the state of things today. Booklet solves this problem in minutes: when you point Booklet.ai at a Sagemaker endpoint, you get a responsive web app (steps below).
Turning your ML Model into a web service is painful
Today, figuring out how to properly expose your
predict function so developers can call it from their apps is far more involved than it should be. For example, if you’re deploying a model via AWS Sagemaker, you’ll need to navigate the labyrinth of the AWS console to setup Lambda, API Gateway, and more. If you get a basic API going, you probably don’t have necessary bits like error monitoring (what if a small change in pre-processing goes wrong?) and performance monitoring in place (what if inference slows to a crawl due to increased volume?).
Booklet solves this problem in minutes: when you point Booklet.ai at a Sagemaker endpoint, you get a production-ready HTTP API for your ML model too (steps below).
It’s too hard getting
predict results in front of the end user
It’s frighteningly easy to put an incredible amount of work into an ML model that doesn’t show
predict results inside the apps that end-users actually view. Anyone can build a lead scoring model, but can you put the results into Salesforce quickly so the sales team can react? Are they sent to Mailchimp as well so folks are added to the correct campaign based on their score? Does a trendline of the average lead score appear on management’s dashboard?
Booklet’s integrations let your predictions flow to the services that need them. For example, you can feed AWS Redshift rows into your model, then send the model predictions back to Redshift, out to Salesforce, Mailchimp and more. If a dataflow fails, Booklet logs it. If it takes longer than normal, Booklet logs it.
A data scientist should spend more time doing what they do best: creating models. A web developer should spend time on unique business features, not building and monitoring ML data flows. Integration work is the plumbing of ML model deployment: it’s involved work that no one really enjoys but everyone needs.
Booklet is the web app your ML model deserves
A modern web app does many things: it has a responsive UI. An HTTP API. It integrates with other services. It’s monitored for errors and performance issues. We built Booklet.ai to be the easy button for creating a web app on top of an ML model.
How it works
Booklet sets up a testing UI and easy integrations in minutes. All it takes is 4 simple steps. For a more detailed tutorial, from end to end, check out our lead scoring example.
1. Sign Up for Booklet.ai
Your first model is free on Booklet! Sign up below:
2. Connect your AWS Account
Input your AWS Sagemaker (support for more tools coming soon!) credentials and select your endpoint.
3. Setup the UI
With a few simple configurations, you can now easily test and share a UI for easily testing your model.
4. Setup a Dataflow
Setup a source for model inputs (AWS Redshift or another data store) and one or many destinations (Sales or Marketing Tools) for where to send the model outputs. Now you can kickoff a dataflow to automatically send model results to places where they will be useful to the business.