Genetic Algorithm as a Service: A Lucrative AI SaaS Business Idea.
Thinking outside the box on what an AI Business Is
I am absolutely sick of hearing about everybody’s cookie cutter SaaS ideas. Seriously, it feels like everybody’s idea of a company is “have ChatGPT learn from your own internal data”. If there’s one thing that software engineers and technical founders lack, is creativity.
So here I am to present 2 lucrative ideas that I promise you have never heard before. How do I know this? Because I Googled it, and not one relevant result came up. I’m writing this article for several reasons. For one, I want to prove that your business idea does not matter. Even if I tell the entire world (or all of Medium in this case) about my lucrative business idea, nobody is going to steal it, despite it being an AMAZING idea! The second reason I’m writing this is to solicit user feedback: if you’re a SaaS Founder, would you buy this service?
The Business Ideas: “GoaaS” and “PeaaS”.
If you read the screenshot, then you probably saw 5 words you’ve never seen in the same sentence: “genetic optimization as a service”. 90% of the population probably have no idea what genetic algorithms are and how one could sell it as a service. This article should hopefully convince you about how awesome this idea is! Along the way, I’ll use the example of a second SaaS idea that is powered by the first idea: “prompt engineering as a service”.
For a video summary on these ideas, check out this link.
Quick Recap: What is genetic optimization?
Genetic optimization is a biologically-inspired artificial intelligence algorithm. It’s a search heuristic that mimics the process of natural selection to find optimal solutions to problems. It starts with a set of possible answers and then uses selection, crossover, and mutation to iteratively evolve these answers toward better solutions.
To see an illustrative example of genetic optimization as applied to the financial domain, check out the following article.
Genetic Algorithms (GAs) can be used in other domains outside of finance. For example, it can be used as the backbone for a “prompt engineering as a service” product, as I will discuss in the next section.
Practical Use Case: Prompt Engineering as a Service
Imagine you’re a SaaS company and using AI in many of your internal and external products. Because many of your products are powered by Large Language Models (LLMs), you have a variety of different prompts scattered through your codebase. Iterating, improving, and testing your prompts is extremely difficult, and it often requires long build and deployment cycles. Controlling costs and speed is also nearly impossible, especially at the same time.
Now, imagine the polar opposite scenario. You’re using an external service to upload and manage all of your prompts. Your prompts can be re-used across different services, and can be easily duplicated, edited, and fine-tuned for specific use-cases. The best part is, after you give it an initial prompt, you don’t have to perform any prompt engineering! You simply give the model a rating for all of its output, and the model learns to get better and better over-time. This is because the prompt management system is powered by genetic algorithms!
How would it work?
When thinking about a genetic algorithm, you have to ask yourself “what would I be optimizing exactly?” In the context of a prompt engineering application, you can give the users the ability to optimize any aspect of your prompts. For example, you can seek to minimize costs while maximizing accuracy and user satisfaction. The nice thing about genetic algorithms is that they are very good at optimizing any arbitrary function.
And they say neural networks are universal function approximators…
Next, let’s think about how we would encode a prompt into our genetic algorithm. To do this, we must create an “individual” which is essentially an array of numbers. This array is the object that undergoes selection, crossover, and mutation operations to create new potential solutions.
So how do we encode our prompt into an array of numbers? When doing prompt engineering, what are some things that we can change about the prompt? Remember, most good prompts are composed of the following format.
Role: Who the model is? For example, “you are an AI JSON Generator for a FinTech company.”
Context and Constraints: What is some information about the problem space and what constraints should the model adhere to? For example: “you must always respond in syntactically valid JSON”
Examples: What are some example conversations we could have?
Additionally, we can increase the number of parameters to optimize by introducing Natural Language Processing (NLP) algorithms. For example, we can do stemming, lemmatization, introduce a parameter for maximum length, a parameter for the number of examples, and potentially introduce a Retrieval Augmented Generation (RAG) Pipeline. If we really wanted to get creative, we could optimize the hyperparameters of the RAG pipeline too, but we won’t discuss that now.
The optimization process would be straightforward. An “individual” is just a prompt that underwent different preprocessing steps and used different models (open-source vs GPT-3 vs GPT-4). The objective function would be maximizing the model accuracy or minimizing the response times. And remember, genetic algorithms can optimize multiple functions at the same time.
Over time, you will literally develop the perfect prompt based on your specific use-case! You won’t have to do prompt engineering: the genetic algorithm does it for you! Just give it an initial prompt and let the model slowly run and get better over time. This has the effect of decreasing AI costs and increasing accuracy and response relevancy at the same time when deploying these models.
Packaging it Up
To deploy the app, I would include it as a feature for my prompt management platform NexusGenAI. NexusGenAI is an AI Configuration and prompt management platform. It powers the AI Chat in NexusTrade, and allows users to create unique, configurable automated investing strategies.
The immediate impact is that the AI Chatbot would become smarter and start slowing learning how to provide better responses. Then, costs would also come down, as the prompts would be more streamlined and efficient. And I could slowly build an arsenal of self-improving AI Agents. That’s insane!
Hey You! Are you a business owner or content creator? Would you use either of these products for your business? 🤔 Comment down below!👇🏾
Discussion and Conclusion
If you made it this far in the article, you’re probably asking yourself this question: how do I know this SaaS will be lucrative? I haven’t done product research. I didn’t interview any SaaS founders and ask them if they’d purchase it. But I still have unwavering confidence. Why is that?
To start, users don’t really know what they want. To quote Steve Jobs, “People don’t know what they want until you show it to them.” Nobody is going to understand what a “genetic optimization algorithm” means or why a “prompt engineering platform” is the best thing since sliced bread. You just have to show them. And then they’ll get it.
Secondly and most importantly, the reason why I am confident these SaaS ideas would succeed is because they would provide a lot of value. Almost every business will be deploying AI models for some aspects of their businesses. The ability to manage all of your different models in a centralized place, and have it continually self-improve without prompt engineering? Every single company will want to buy this product; they just won’t know they want it until they see it right in front of them.
When starting this article, I set out to do two things. For one, I hoped to convince you that genetic algorithms as a service is the coolest business idea since ChatGPT 😎! Secondly, I hoped to prove that the business idea itself isn’t worth a dime.
It’s execution that matters.
Thank you for reading! If you enjoyed this article, please give me some claps and share this article with a friend! I have several newsletters you could follow. Aurora’s Insights is the perfect blog if you’re interested in artificial intelligence, machine learning, finance, investing, trading, and the intersection between these disciplines. You can also create a free account on NexusTrade to get access to the state-of-the-art genetic optimization engine for trading strategy optimization.
NexusGenAI also has a blog and is accepting new sign-ups on the waitlist!
🤝 Connect with me on LinkedIn
🐦 Follow me on Twitter
👨💻 Explore my projects on GitHub
📸 Catch me on Instagram
🎵 Dive into my TikTok
