Webinar | AI Assisted Data Science - Tony Stark and Jarvis w Bindu Reddy

Artificial Intelligence, and machine learning, are huge developments. While we may just be at the infancy of their potential, most of us in the industry understand how groundbreaking these technologies are. Artificial intelligence and machine learning provide highly empirical, increasingly accurate, and acceleratingly efficient ways to study the world we live in. This ranges across a broad spectrum of industries, disciplines, and fields of research. The best way that I can categorize the importance of artificial intelligence is this:

It takes a human upwards of 14 years to understand the purpose of triangulation. Artificial intelligence performs mathematical calculations, and computations, almost instantaneously - with substantially higher accuracy, with vastly more calculations performed than a human brain ever could. Put simply, it is our way to be accurate.

What takes us up to14 years to learn can be handled, in several degrees higher orders magnitude, by artificial intelligence nearly instantaneously.

This article summarizes my takeaways from the Abacus AI service provisions, and a high level overview of the types of problems a product like this could assist with, and speculates ways Acorn Interactive Inc. could be a partner in configuring and deploying this technology for industry partners.

Meet Bindu Reddy

Oh hello, Bindu!

I came across Bindu on twitter (@bindureddy) because she sounded like someone I could actually talk to about career interests. She openly discusses many of the the ways I would imagine schemas and database models working to their optimum potential, with automation and inference - in the wild. I learned about her during the Covid-19 pandemic on twitter, and it's been an amazing experience to learn from her online. She has a breadth of expertise in a highly important field, and a track record to support it. As the CEO of Abacus AI, it's important to say she's a hugely interesting follow and, because I have a chance, to acknowledge that I am a HUGE fan.

Bio from website:

Bindu Reddy is the CEO and Co-Founder of Abacus.AI. Before starting Abacus.AI, she was the General Manager for AI Verticals at AWS, AI. Her organization created and launched Amazon Personalize and Amazon Forecast, the first of their kind AI services that enable organizations to create custom deep-learning models easily. Prior to that, she was the CEO and co-founder of Post Intelligence, a deep-learning company that created services for social media influencers that was acquired by Uber. Bindu was previously at Google, where she was the Head of Product for Google Apps, including Docs, Spreadsheets, Slides, Sites and Blogger. Bindu has a Masters Degree from Dartmouth and B.Tech degree from Indian Institute of Technology, Mumbai.

I also really admire the types of encouragement she shares towards her followers to get engaged with the science and mathematics behind machine learning. Regularly she reminds her audience about the prerequisites it takes to excel in this field, and leaves specific examples of the kinds of maths people should strive to learn, in order to become effective data scientists.

(passing it on)

Personally, I am using Khan Academy to keep current with the knowledge gaps and will be getting to statistics eventually.

­čĄô┬áthis is what makes twitter worth tweeting, Bindu.

So, where were we... ah yes:

a slide from the presentation

As operations grow, things get more complex.

Think about it for a moment. You want to fly to the tropics. You NEED that vacation. But you are one of 9 billion people who want to do the same thing. So what are the steps?

  1. You book a flight
  2. You take a cab to the airport
  3. You scan your boarding pass
  4. You drop off your luggage
  5. You go get a premium soda water
  6. You are notified the plane is boarding somehow
  7. You spend a couple hours in the stratosphere and marvel at the wonders of aviation and engineering.. right?

Now think about steps 1-7 and multiply that same operation for 9 billion people and never screw up a single step...

This is why we need artificial intelligence assistance, across a multitude of areas.

Artificial Intelligence helps to make sure empirical accuracy is applied to each step of the process.

Every step listed above has, or at some point has had, an entire PhD dedicated to its study. Booking requires measurements of capacity, time and scheduling, and needs to correspond with an actual aircraft. Cab services need to triage cab request volumes, availability of vehicles, the locations of operators, and estimations of departure and arrival times. A boarding pass needs to be about your flight, not Bill's trip to Estonia next week.

Quick side note it has been argued that NFT's are a great solution for aviation ticketing (link)...

The flight attendant needs to relay the correct flight information, your baggage needs to land in the correct destination, and... well a plane also needs to work and have fuel.

You get it, there is so much going on. Trying to grab a sheet of graph paper and figure this all out manually, with accuracy will more often than not lead to catastrophe. We need assistance with every step of the way, and we need a unique blend of metrics, data, and estimation to satisfy our needs for quality customer satisfaction*.

Having established a need to invoke these tools, techniques, systems, and protocols we can move into the what, how and the who part.

*Please note, I say customer satisfaction in the context of aviation and travel. The scope of AI extends well beyond commerce - I am just keeping it to a relatable example.

trade terminology - there is so much of it out there, sometimes these reminders are essential.

Definitions

Straight up, I go to these events to learn and get better. I encourage and resource my team to attend them for their domains of expertise. One of the most important skills to develop from attending these events is self study on the terminology and verbiage. It's super important to get on the same page for commonly used terms in this field. If this is familiar to some, please just skip over this section.

One of the services Acorn's blog serves is just relaying industry terminology, and insights. We take time to attend these events so we can facilitate better relationships, to build better solutions.

Machine Learning

From Merriam Webster Dictionary (link)

the process by which a computer is able to improve its own performance (as in analyzing image files) by continuously incorporating new data into an existing statistical model

Neural Network

From Merriam Webster Dictionary (link)

a computer architecture in which a number of processors are interconnected in a manner suggestive of the connections between neurons in a human brain and which is able to learn by a process of trial and error

Feedback Loops

From Devin Soni, on Medium (link)

In machine learning systems, we often receive some kind of feedback from the modelÔÇÖs environment, that is then fed back into the system. This can take many forms, such as using the modelÔÇÖs output to train newer versions of the model, or using user feedback on the modelÔÇÖs decisions to improve the model.

Generative AI

From MLQ.ai (link)

Generative AI is a sub-field of machine learning that involves generating new data or content based on a given set of input data. This can include generating text, images, code, or any other type of data. Generative AI typically uses deep learning algorithms, such as generative adversarial networks (GANs), to learn patterns and features in a given dataset, and then generate new data based on the underlying input data. The goal of generative AI is thus to create new, synthetic data that is as close as possible to the real-world input data. The most common example of generative AI right now is of course ChatGPT, which is a Large Language Model (LLM) that has taken the internet by storm and crossed 1 million users in its first 5 days...

End to End MLOps

From Microsoft (link)

MLOps is based on DevOps principles and practices that increase the efficiency of workflows. Examples include continuous integration, delivery, and deployment. MLOps applies these principles to the machine learning process, with the goal of:Faster experimentation and development of models.Faster deployment of models into production.Quality assurance and end-to-end lineage tracking.

Pipelines

from seldon (link)

Machine learning pipelines are used to optimise and automate the end-to-end workflow of a machine learning model. Core elements of the machine learning process can be refined or automated once mapped within a machine learning pipeline. As more and more organisations leverage the power of machine learning, models are developed and deployed within increasingly diverse settings. A machine learning pipeline helps to standardise this development, helping to build efficiency and model accuracy.

Great, super straightforward stuff ­čĄ¬┬á

Also, reinforcing the learning can happen in any number of ways. Because the trade terminology is, to most people, plain alien... have some fun with the fact we're all beginners of sorts. Try to find ways to practice and rehearse these concepts. This could be through coding, study, or constructive engagement with these professionals on social media.

another slide outlining Abacus's capabilities

Let's put the themes together now.

On the one hand you have the inevitability of multi layered, repeat use and ungodly logistically complex suites of operations. On the other you have highly specialized, technology dense, and computationally determined tools to ensure the unweildly complexity of the operation doesn't, for lack of a better word or high level context...

go bananas.

Systems with a lot of moving parts, interacting with living breathing humans, or complex instrumentations of machinery, should not go bananas.

That is about the simplest way to explain the importance of artificial intelligence and machine learning.

My null value or undefined goodness...

Of course saying that one needs to

Ôťî´ŞĆdo the A.IÔťî´ŞĆ

is in almost every context not going to cut it. A team of trained specialists need to harmonize their skills with a comprehensive and ever growing list of business demands. Organizational operators need to have a grounded sense of what they can do, what they need to do better, and where they would like their operations to arrive in the future.

In this sense, one could consider the role of MLOps and data scientist teams as a proverbial tango-dance-partner with a suite of objectives. Objectives that need to be

  1. communicated
  2. understood
  3. measured
  4. analyzed
  5. reviewed

And then steps 1-5 over, and over... and then over again.

This is why instrumenting a sound plan, or in technical terms, model is fundamental to success in this space. Once a model can be established to encapsulate the operation, or rather, a very specific part OF the operation that needs to be measured, machine learning can run those instructions over and over, and over, to produce new and novel insights.

More terminology - A Model:

from Microsoft (link)

A machine learning model is a file that has been trained to recognize certain types of patterns. You train a model over a set of data, providing it an algorithm that it can use to reason over and learn from those data. Once you have trained the model, you can use it to reason over data that it hasn't seen before, and make predictions about those data. For example, let's say you want to build an application that can recognize a user's emotions based on their facial expressions. You can train a model by providing it with images of faces that are each tagged with a certain emotion, and then you can use that model in an application that can recognize any user's emotion. See the  sample for an example of such an application.
Kind of like this but for analysis and insight generation. It's science.

Get it? A data model... apparently Gen Z doesn't even watch Zoolander. Whatver.

At this point I am going to imagine many people are thinking: this sounds like a lot of work...

And it is! This is exactly why organizations like Abacus AI exist. Licking ones finger and sticking it into the wind is no way to certify oneself in the fields of artificial intelligence and data science. Having robust service providers unpacking the specifics and provisioning both tools and expertise to administer solutions here can save companies a ton of money and guesswork.

Having specialists to reach out to, to provide platforms to start with the complex operations of machine learning is one way we can put together complex workflows with predictability, the right personnel, for amazing outcomes.

I'm sure you might also be wondering how Acorn Interactive Inc. may fit into this fold.

Like anything, it starts with having a conversation.

As some of you may know, we have recently launched our core services page. A huge part of our internal process is to develop well designed, highly iterative, resources to help explain the types of work a service provider like ourselves will perform for the companies we seek to partner with. As such, what on the surface may appear as marketing materials (which they are) have substantially more devils-in-the-details. Well, not devils ... no. Angels, sure... better ­čĄö┬á

Referring back to this quote:

End to End MLOps

From Microsoft (link)

MLOps is based on DevOps principles and practices that increase the efficiency of workflows. Examples include continuous integration, delivery, and deployment. MLOps applies these principles to the machine learning process, with the goal of:Faster experimentation and development of models.Faster deployment of models into production.Quality assurance and end-to-end lineage tracking.

Acorn has been building out its devops capabilities to be able to leapfrog into the next chapter of our enterprise. The way in which we do this is to publicly demonstrate how our suite of workflows will "merge" into more advanced, and automated ones with the right mixture of investments, revenue, and above all - skills development.

(also why NOT plug on your own blog. anyway)

Let's do a thought experiment on how we could begin to automate with Abacus.ai

In the development of our services pages, Acorn worked closely with our development and design team to ensure that we had a clean process for our DevOps.

Please note that we do have a code/syntax highlighter feature coming but a "raster image" will have to suffice until it is ready.

A draft of the services page model.

Model drafting, then, was our way to have the team familiarized with the steps demanded by an MLops outfit. For now, our model only incorporates content provisions. This includes

  1. The content outlined on the Services Listing Page and
  2. The content outlined on the Services Single Page´╗┐

´╗┐´╗┐What Acorn is doing, then, is architecting the data models to do... whatever the services pages require. This is a common theme that happens in our industry - people expect designs to be outstanding (which they should), but the omit the complexity of due diligence on back end related tasks, and workflows. This stage, of beginning with the model and schema properties for a given "web page" allow applications to have greater degrees of granularity as they evolve into more complex operations.

Saying build the model and they will come is a bit misleading. It is more like build the model and you will have sanity in your database.

Voila! We have a database model and have stored data to it.

The next step for us is to ask the operational question of:

Great, now what the heck do we do with this?

We have our company LinkedIn presence and we have our twitter account.

All social media platforms have what are known as marketing API's:

  1. ´╗┐LinkedIn API | Marketing´╗┐
  2. ´╗┐Twitter Marketing API ´╗┐

´╗┐´╗┐The most important thing Acorn can do right now, is get started on these platforms. Since we have a "schema" to house the services page data, we can iterate on this "model" to run api requests TO any marketing api.

This is how we would expect to start using Abacus for our internal operations.

There are a lot of ways to gain insights from data, but we need to have clear goals established to operate with a service like Bindu's. Do we:

  1. Want to measure and optimize the performance of a given website page across a multitude of social networks?
  2. Compare the different service offerings and see which one gets the most traction for the most amount of revenue?
  3. Simultaneously evaluate the cashflow of our Quickbooks account with our ad spend within a given period?

It is very important to have clearly defined goals established prior to the instantiation of an AI service like Abacus, or else we run the risk of summoning the "code spaghetti monster"

all fear the poorly structured code spaghetti monster.

Let's recap:

  1. We have a database schema for our services
  2. We want to be marketing our services online
  3. We need to know which services are in demand from customers
  4. We don't know which ones will get traction
  5. There are tools for aggregating data about this
  6. They cost money we need to measure
  7. To become financially sustainable we want to action insights that help us to sell our core service offering and to do that
  8. We would need to amend our first version of the service schemas to house reports from a variety of marketing platforms
our search console report
words from our indexed site that get shown on a google search

Oh, and before I forget - we'll also need to figure out how the verbiage indexing informs the types of visitors we get from our search engines, and how all of that interoperates with our multi-social platform approach to sales. Hopefully the message is starting to resonate.

There is a lot of coordination that goes into effective data wrangling, and at the end of the day AI and Machine Learning are there to help.

This does get very confusing, and only worse without measurement and planning. As Bindu mentioned in her presentation, on of the key roles of data scientists is to make adjustments based on observations from reporting, and more specifically, data visualization.

What Acorn is building is way to start to do the data visualization, in order to generate insights for sales. In doing so, we plan to showcase how to configure systems like Abacus, not just for ourselves, but the customers we serve.

It takes a lot of communication, time, and teamwork to get these things right. Moving forward, we also will be releasing a jobs page for our business, where we will be recruiting the types of programmers who can assist in building out these data pipelines.

The starting point, then, is to have a stable model in our devops protocols with which we can extend to working with an amazing system like Abacus.

All we need now is for the stars to align, just a little bit. The best way for this to happen - just like administering a sound model and feedback loop configuration - is to write down the details and correspond with professionals.

Thanks for the talk, Bindu!