Why I'm building Optimizely for AI Prompts

Hey folks

This is going to be a quick little story about my Founder journey and how I got to the point that I'm at now, working on a two projects and living and breathing nothing but AI for the past 4 months.

A Little Background

It all started off on YC's Meet a CoFounder page. There I was able to meet my current CoFounder Lukas. He had an idea, and I was all out of business ideas. Despite that, I had about 6 years in the software industry across multiple companies and had been a self taught web developer since I was 12 years old.

Lukas and I hit it off, and we began building our first project together. OnestaFinance was a marketplace for Retail Investors to Meet, Rate and Review Financial Advisors and RIA firms. The idea was to bring Honesty to the industry of Finance (hence Onesta, the root word for Honest)

Building OnestaFinance

I forgot to mention that at this point we had a 3rd technical CoFounder, but he ended up dropping out of the project later on due to personal reasons.

I had a really fun SEO strategy for OnestaFinance. Basically, we had scraped every single financial advisor and advisor firm's information (about 400,000 people's public data) and created profiles for everyone. We then added all 400,000+ profiles to a sitemap, filled them with SEO keywords on that person and let google start indexing.

We had massive amounts of early traffic from this, but unfortunately it wasn't traffic that we knew how to monetize so we built this without receiving a drip of revenue. I think at our peak we hit around 20,000 search clicks a month, but I don't have access to that chart anymore because we merged OnestaFinance into our new project.

The Pivot

onesta.ai

The story behind onesta-ai was, while we were working on OnestaFinance we realized that building a marketplace is really tough. It's even tougher when one side of the marketplace (the financial advisors) hate you and want their public data scrubbed off your site.

The calls with folks screaming at us wore us down, and we made practically no money on the project. We decided there needed to be a better product we could build.

At the same time, ChatGPT was just released. And we saw the world in an uproar for ChatGPT enabled tools. We spent about a month or so experimenting with ChatGPT, Langchain, HuggingFace and more. We eventually had the idea for onesta-ai, an AI enabled sales email tool. Think Mailchimp but with auto generated content personalized for each contact instead of templates.

Present Day

Jumping to the present, I've been building onesta-ai for the past 3 or 4 months (time starts to blend together as a founder so I'm not sure). We've been working hard on improving our prompts and I realized, I have prompts across 3 separate code repositories that I need to update. It sucked, especially trying to test my prompts beforehand.

This gave me another great idea, why not make a tool like Optimizely but for prompts. We could have a quick, reliable prompt retrieval. We could have a dashboard that allows prompt engineers to spin up and test new prompts on the fly. And the only code changes required are hooking up the prompt strings to the results of these API calls.

RPM (Remote Prompt Manager)

Thus, RPM was born.

Check it out here
https://github.com/AgentBurgundy/rpm

RPM stands for Remote Prompt Manager, and it's a Node.js service that does exactly that. It remotely manages your AI prompts.

I just started building it quite literally last night, so let me tell you where it's at now and where it will be at in a few months.

Right now it is a service that can be hosted by anyone. It takes a little setup but I hope to have a managed version out soon that will make it easy for anyone to spin up an instance of RPM.

It uses a MongoDB and Redis instance that must be configured in the .env for the service so it knows how to connect to them. The MongoDB stores the prompts you built and the Redis instance caches results to try and make repeat responses faster.

Where will it be in 3 months

In 3 months we will have a dashboard and Node.js SDK to make using the service even easier.

We will also have a hosted version where you can spin up your own instances cheaply.

In three months, I fully expect to have about 25 - 50 contributors to the project, making RPM the best possible solution for managing your prompts remotely.