Lately, I’ve been juggling several ideas for starting businesses around software. With AI accelerating, global weather events increasing, and a tumultuous political landscape, I feel the pressure to find greater security. This pressure often makes me question whether I’m using my time wisely. I must give my current garden time to grow, but also not ignore fertile soil.
For now, I’m settling on the thought that I haven’t dedicated myself to one project long enough — and that I should continue refining my idea for an AI personal trainer and workout logger app.
Recently, my areas of focus have been developing my iOS app Hybra, learning about RAG and vector databases, and running large language models on private networks.
iOS
The iOS development has been going well. I started mobile app development on May 22, 2025, and deployed my first version to the App Store on August 1, 2025. My goal is to create a workout logging app that reduces friction when recording workouts. I currently use the app in two ways:
- First, by recording exercises as I go through a workout with no set plan — freestyle.
- Second, by planning the workout ahead of time, either manually entering exercises or using the AI feature to generate workouts.
Check it out: https://www.trainhybra.com
LLMs
Over the last month, I experimented with setting up a Python script that let me chat with an LLM via voice-to-text. After realizing I wanted to focus on the AI components, I scrapped the voice aspect. Chatting with a locally running LLM was just as convenient and reduced overhead significantly.
After reverting back to a text-based chatbot in the CLI, I added a RAG layer that analyzed documents and injected relevant information into prompts delivered to the LLM. This was a success — I could get accurate responses.
Deploying LLMs
After experimenting locally, I wanted to figure out how to set up one of these systems online. I explored Cloudflare’s offerings and decided to use one of their hosted LLMs with a RAG database service to create a chatbot enhanced by custom information.
I was able to deploy a server and UI in an evening. The ability to launch projects to the web has gotten serious upgrades in the last few years. At this time, https://www.nash-chat.com is live and can answer Nashville-related questions. Having experience building with Cloudflare products made the process smooth and abstracted away the complexities of working with vector DBs.
Applying Lessons
Now that I have some experience with RAG and vector databases, I can imagine new ways to enhance my personal trainer AI. Maybe users could provide their own training data to improve interactions? Or I could supply specific rules and information for training toward particular sporting events. Not quite sure yet.
I want to create a tool that people would be willing to pay for. While workout generation and insights are interesting, I’m still unsure if that alone is enough to keep users engaged long-term.
Next Steps
My next step is to ensure my onboarding flow is working robustly. Then, I’ll move on to some marketing and begin implementing a paid subscription with RevenueCat:
https://revenuecat-shipaton-2025.devpost.com
They’re hosting a hackathon this month — it’s the perfect opportunity to get motivated, find an audience, and deploy new features.
I’ll report back.
Cheers!