Gravity AI’s Top 3 Production Ready Models

We’re always looking for ways to improve and streamline our day-to-day operations, but sitting down to write complex code can take up precious company time and resources. Implementing a production ready model can not only save you the man-hours it would take to build a model yourself, but also give you the ability to integrate the model through some easy to use API endpoints. But wait, what does "Production Ready Models" mean? Put quite simply, it means this AI model is ready for production. The models have been pre-trained on a vertical/industry specific set of data. The custom gravityAI Docker container is ready to be deployed into a production environment through a process of API endpoints. My dudes, we set out to allow you to plug the model into your pipeline as a microservice, ready for you to put your data in and get those sweet, sweet results out.

We have a few favorites right now (that's right, we play favorites, wanna fight about it?), that both us and our clients see as awesome and valuable. So here's our picks for the top three Production Ready Models that Gravity AI offers and some scenarios where they can be best utilized.

Text interrogator  

Yes, we actually named this Question/Answer Model the Text Interrogator. Are you tired of slogging through large documents for hours looking for the information you need? We’ve got your solution. Our Text Interrogator Model allows you to take the unstructured text in a long document — such as a financial report, purchase order, or contract — and extract relevant information using natural language questions. This model is great to drop into a pipeline of document workflow automation.

The Text Interrogator is a pre-trained question-answering language model based on Google Brain’s Bidirectional Encoder Representations from Transformers (BERT) language model. It’s been trained using the Stanford Question Answering Dataset (SQUAD) (among many other proprietary data sets) and tuned over the span of two years to provide some pretty stellar results for digitizing financial documents. The model is structured to take a query document in the form of plain text and a natural language query as input and provides an answer to the natural language question as output. 

On aside, it's really fun to feed it text from A tale of Two Cities and go nuts.

Real use cases for this model include:

- Extracting data from unstructured invoices.

- Saving hours and resources to extract information from long contracts.

- Automating extraction of HR forms/paystubs.

Topic Similarity Model

Wish you had a way to dig through massive amounts of text to find hidden sections related to a topic? Crankset Analytics’s Topic Similarity Model does just that. This container compares the text from a desired topic against a database of docs/text, surfaces individual sections that are related to that topic, and provides you with a similarity score for each section. 

It’s important to note that this model ain’t your average search program. Instead of hitting Ctrl-F to look for the same phrased information, this model pulls out any relevant bits of information that could potentially tie into your original topic, combing through pages of large reports or research so you don’t have to. 

Now we just need Elon to pick up the speed on Neuralink so I can connect this model to my brain and gain nerdy, academic super powers.

Use cases for this model include:

- This model was originally built for the government to dig through massive amounts of research.

- Creating an app for students looking to dig deeper into a database of research topics.

- Reducing large sets of unknown documents into smaller, more manageable sets of related topics.

- Prioritizing and enhancing unstructured data topics into structured values.

V.E.R.N Sadness, Anger, and Humor Emotion Detector Models

Keeping tabs on people’s moods online is an important part of a company’s internet presence. These three V.E.R.N. Emotion Detector Models (Sadness, Anger, and Humor) interpret the complex emotions layered into a text post online. Unlike a regular sentiment analysis program, these containers go beyond identifying a post as “positive, negative, or neutral.” These models will detect complex human emotions (going as far as to even identify sarcasm) and express their findings as a confidence percentage in real-time.

Most useful for nuance communication or detection, these models can be used separately or in conjunction with one another. 

Use cases for these models include:

Identifying customer satisfaction in a customer service chatbot.

Monitoring the mental health status of an at-risk patient.

Identifying how your company’s investors speak about your finances.

Hear that gravityAI investors? I'm monitoring your emotions on us. Got my eye on you. Feeling drawn to one of these models? It’s probably Gravity.

Any of these Production Ready Models are ready for implementation through a one-time, easy purchase with flat monthly license fees. No sending us any of your sensitive data, no working around our database — it just takes one quick click for your container to download and begin working for you in less than a day. Easy, right? After you purchase a model, you can retain access to its license as long as your subscription is active. Whether you’ve got a lot of code or absolutely none at all, these containers will fit seamlessly into your standing infrastructure. 

Purchasing a model means you pay one flat monthly fee and can utilize as much data through your container as you want! It’s predictable, with no surprise spikes for your procurement or bulling team to look out for.  

Ready to sign up for your demo? Click here to get started!