Iterate's AI Patent #1

Drag-n-Drop AI -- useful patent #1

Iterate has been granted 4 patents so far. We have 20 more pending or in draft.

This post explains the 1st of the 4. Lawyers call it Container architecture for modular machine learning. Internally, we call it Drag-n-Drop AI.

Think about that.

It could also be called Super-Fast to AI. Or Speed-to-market AI. Or Lower TCO AI, meaning total-cost-of-ownership AI.

Below, are 3 ways to look at this Drag-n-Drop AI patent.

1. the facts (grant date, status, inventors, assignee, etc.)
2. the "summary" submitted to the patent office, and
3. a layman's look at a useful AI patent. This may be of most interest to most people.

Here are 3 ways to look at AI patent #1:

1. The Facts

Name: Container architecture for modular machine learning

Patent number: 10776686

Type:?Grant

Filed: Nov 22, 2019

Date of Patent: Sep 15, 2020

Assignee:?ITERATE STUDIO, INC.?(a.k.a. Iterate.ai)

Inventors:?Shomron Jacob?(San Jose, CA),?Brian Sathianathan?(Morgan Hill, CA),?Arulkumaran Chandrasekaran?(Tamilnadu),?John Selvadurai?(San Jose, CA),?Chatura Samarasinghe?(San Jose, CA),?David Ruvalcaba Gonzalez?(San Jose, CA)

Primary Examiner:?Alan Chen

2. Legal SUMMARY submitted to the U.S. patent office:

The present disclosure relates generally to systems and methods that include a configuration server with a processing element operable to operate a plurality of containers. A plurality of functionality modules are operable to execute within respective containers of the plurality of containers. Each functionality module includes: an input, a functionality operable to be executed by the processing element to perform an artificial intelligence operation using the input, and an output produced by the functionality. The modular system includes a messaging object operable to connect the output of a first one of the plurality of functionality modules to the input of a second of the plurality of functionality modules. The modular system configures the messaging object to adapt the output of first functionality module to the input of the second functionality module. (The full patent can be read where it's published on the U.S. Patent Office's website.)

3. In layperson terms, this is what patent?10776686?says:

Again, inside Iterate.ai we call this patent Drag-n-Drop AI.

This patent is all about a new way to build and manage artificial intelligence (AI) and machine learning (ML) applications. Traditionally, you'd need a really smart data science engineer to build these kinds of things, and they'd have to keep babysitting the application even after it's done.

Not ideal, right? You'd need an expert at every step of the way, and these experts aren't exactly standing around on every street corner.

So, what's the solution? This patent says, "Hey, let's make this whole process modular." Imagine it like building blocks. Each block is a separate module that does its own thing. They're all running in their own little worlds, called "containers."

Each module has an input (where it gets its data), a function (what it does with the data), and an output (the result of its function). The beauty of this system is that these modules can be connected together. The output of one module can be the input of another, kind of like a relay race.

But how do we make sure that the baton (or data) gets passed smoothly from one module to the next? That's where a "messaging object" comes into play. This object makes sure that the output of one module fits perfectly into the input of the next module. And one or more of these modules can perform AI operations.

Now, the whole system becomes way easier to manage. Need to add a new function? Just add a new module. Need to change how the data flows? Just change how the modules are connected. It's like a Lego set for AI applications! This is the Drag-n-Drop approach to creating custom GenAI or AI experiences:

And the cool thing is that it's not just for building AI applications. It can also be used to manage them in real-time. Say a user makes a request. This request is processed by a series of modules, each one doing its own thing, until finally, a response is generated. This response is then translated back to the user.

So, in essence, this patent is trying to simplify the whole process of creating, managing, and executing AI and ML applications by making them modular and easier to handle. This can potentially open up AI and ML application development to a broader range of people and not just the data science pros.

The Yellow Blocks are hard-core AI Blocks (a.k.a. Containers)

Above and below are pictures of our Interplay platform. A yellow node -- the color -- indicates that it's an AI Container. Yellow = AI in Interplay. Some of our ready-made AI capabilities include:

• Data Analysis: Regression, Classification, Clustering
• Computer Vision: Image Recognition, Object Detection, OCR
• Generative AI: Text Generation/Summarization, Image Generation, Multimodal Systems

Other colors represent categories like Enterprise Connectors (SAP, Oracle, Salesforce, etc.), Commerce Containers (Braintree, etc.), and so on. Orange is the color for Enterprise nodes.

There are more than 500 Containers custom built into the Interplay platform, with about 4,000 available through the Open Source community.

Below is an example of a flow we use for license plate recognition, done at high speeds, using a Large Language Model (LLM) in a large company's private environment. Our response times are in milliseconds on a GPU (0.69 seconds), which a C-Store or Gas Station would tell you is very important for a pay-with-plate (license plate) solution. We get a 1.9-second response time running on a CPU.

The system flow can be found on our website, but this graphic shows a partial process. Interplay processes the AI on the edge (meaning at the local store) instead of sending the AI request to the cloud.

License plate reading can be useful in a wide range of situations -- from gas stations to schools to restaurant drive-throughs to grocery click-and-collects.

Drag-n-Drop can speed up the software development process by up to 17 times -- between 2X and 17X usually. Meaning the total cost of ownership for each solution decreases. You can run more experiments. You can switch things up faster, thanks to the modularity, if need be.

The platform can read license plates, count hot dogs for inventory management, produce graphs for data scientists, scan loan documents for text extraction and classification -- and so much more. It's extremely versatile. Today, it's used by beauty companies, global banks, retailers, convenience stores, entertainment companies, and insurance companies.

Next Up?

Speed is what Iterate's patent #2 is about. Lawyers call this patent a Developer independent resource-based multithreading module.? In one of the next newsletters, we'll describe this one in layman's terms, too.