Nowadays it seems you can't scroll a screen on LinkedIn without bumping into some new Generative AI startup.

While the applications seem endless to an end-user, the core functionality of Generative AI has changed little since ChatGPT first blessed our internet browsers.

These are:

• Data Extraction (transforming unstructured data or images into structured data)
• Data Classification (labeling text or images arbitrarily or strictly)
• Information Retrieval (particularly through RAG)
• Text Generation
• Image or Video Generation

While other functionalities exist, the above handful of features make up the vast majority of Generative AI tools.

• Customer support agents? RAG-based Information Retrieval
• Invoice or contract parsing? Data Extraction and/or Classification
• Sales/Marketing lead enrichment? Data Extraction and/or Classification
• Email customization? Text Generation and/or Data Extraction
• Image classification for medical purposes? Data Classification
• Legal research tools? RAG-based Information Retrieval

Getting ahead with Generative AI

You're the CEO of a marketing software company. Your head of sales frantically runs into your room in a panic about how your top 3 competitors have implemented a Generative AI functionality that provides an enrichment feature to automatically pull key company information from a company's home page.

You've lost a couple of customers, and the world seems to be ending (at least according to your head of sales).

What do you do?

You could:

• Acquire a company that offers this functionality
• Develop the functionality in-house

You take a look at the market and notice people want 50x multiples for their OpenAI wrapper. Your development background finds this unreasonable because the functionality is little more than:

• Scraping a website URL (or URLs if you want to be fancy)
• Generating a prompt to extract structured data
• Saving the data in a database of your choosing

It should be easy right? You run to your development team only to remember that:

• Nobody has worked with Generative AI and nobody wants to learn
• Your code is a monolith where every change has to go through a strict development process, preventing you from quickly iterating on your solution
• There is a drought of Generative AI talent on the market
• You haven't budgeted for an in-house team

You decide you can't let it go and choose to move forward with a dedicated team.

The feature ships in 6 months at best.

The holding company dilemma

Let's change the narrative a bit. You're now the Managing Partner of a vertical software holding company with 20 companies across various verticals. You've got 10 accounting software companies and 10 legal software companies.

You decide to hold a conference for your companies to brainstorm Generative AI solutions that could protect your market position and grow your companies. The event is a success with a myriad of solutions identified by your portfolio companies.

In particular, your companies want to develop:

• A customer support chatbot for all of your companies
• A tool to help with legal research and extract key clauses from legal contracts
• A tool to parse invoices and classify income statement items

What do you do? You're a decentralized holding company so the natural approach for you is to support the initiatives of your portfolio companies (albeit within their approved initiative budgets) by enabling them to build these tools in-house.

But are you really going to:

• Build 20 chatbots?
• Build 10 legal data extraction and research tools?
• Build 10 accounting data extraction and data classification tools?

You could. But then you'd be in the same boat as the simple scenario, except x10.

Doesn't sound like a good time. Especially because you need to show an ROI on each development.

How many development hours will be wasted just on learning the basics?

Generative AI API solutions

At this point, you've likely spent a few months on these Generative AI solutions. It's been getting difficult to keep track of everything and while this has been going on:

• New models have launched that are better at specific tasks
• New features have been shipped that simplify various functionalities
• Your competitors have shipped even more features because they're microservices and you're monolithic

What now?

Why not break down the products into their core functionality and build the same reusable infrastructure for every product?

This means:

• 1 chatbot (copied and adjusted 20 times)
• 1 data extraction tool (copied and adjusted 20 times)
• 1 information retrieval tool (copied and adjusted 10 times)
• 1 data classification tool (copied and adjusted 10 times)

What if in 6 months you could have not one, but all 20 of these tools?

Our Generative AI API solutions:

• Provide a custom API endpoint that can receive any input, perform any operation, and provide the output as JSON object
  • All you need to do is call the endpoint. That's it.
• Do not include royalties or revenue sharing agreements

‍

We also:

• Handle the infrastructure, including scaling, rate limit requirements, and vector databases (for RAG)
• Perform any necessary prompt engineering, model selection, model tuning
• Integrate additional APIs or functionality needed for your tool
• Stay on top of the ever-changing Generative AI landscape to ensure your solution remains cutting-edge