The ProMind story looking backwards
I seek out simple and scalable solutions to various challenges with tech
ProMind for me started from scratching my own itch. When ChatGPT launched, I was a huge fan! One problem I had, however, was having to repeat myself, giving it the needed context on who I am, what I am working on, and my problem space in general. Don't get me started on the prompt engineering madness as well! I hated this! So I set out to build my own.
The initial version was super simple with one value proposition – a collection of personalized assistants each specialized in one specific professional task e.g., copywriting, coding, marketing, etc., but with a twist. ProMind has the added ability to retain context between conversations. You tell it once and it remembers forever, unless of course you delete the message/context, and then it's gone forever as well. Also, none of that prompt engineering struggle. You can converse naturally like you would with a friend. Now, ProMind has grown with various features including image generation, support for plotting graphs, constructing diagrams, sending voice notes, PDF and image Q&A, creating your own assistants, to name a few features.
Over the last 1+ year, ProMind has reached over 130k signups. However, it hasn't been all rosy. With rapid growth comes great responsibility, or something along those lines haha... Within the space of 4 months, a product that started as a project for personal use had gained over 10k signups with over 2k daily active users. This was also the point where I monetized the app, which was doing ~3k GBP in one month. Being the adventurer I am, most of ProMind's embedding models, DB, and entire infra were custom-built and designed for personal use. This marked the beginning of a rapid downturn. Lots of incidents, issues, and slow response times spanning multiple days and at one point, over a week! This was quickly followed by a decline in daily active and paying users and churn. I couldn't dedicate the needed time to properly architect and build the app alongside a demanding 9-5. The AI market is a fiercely competitive one where slip-ups are not accommodated. I learned this the hard way!
Where are we now? Building right back up! The infra is now more robust with monitoring and observability built in. Incidents happen rarely, and when they do occur, they are typically resolved in less than 15 minutes. This was all achieved over an early summer break of intense coding and lots of pineapple juice! More features have also been rolling out on weekends and evenings in-between haha. Also, ProMind now operates on a subscription model!
So far, it seems like I am on the right track. Don't take my word for it. See feedback from one of ProMind's users:
"I'd like to thank you for your product again, as it helped me to win the court trial against the customs."
It's been an exciting journey thus far and one with lots of learning and growth personally! Two lessons that hit home for me are:
Rapid user acquisition is exciting, but without a scalable infrastructure to support it, it can quickly become your biggest liability.
User trust is fragile and hard-won - consistent performance and reliability are non-negotiable for long-term success.
Over the coming months, the goal is to create a one-stop shop for all things professional tasks with AI.
For the growth and tech nerds out there, some recovery stats so far:
Current DAUs: ~200
Paying users: ~40
Current Tech stack: Ionic – Cross-platform web and mobile, NodeJS/Honey (Custom built library for declarative REST APIs - https://github.com/chinaza/honey) – All things non-AI, Python/FastAPI – All things AI, self-managed Docker swarm + EC2 instances, Cloudflare tunnels + on-prem server running AI models and Airflow (I am that crazy haha!), Replicate, Bedrock, OpenAI – some more AI models, Postgres – sensitive data, Mongo – Logging, Firebase – Authentication/Remote Config/Streaming, Sentry, Microsoft Clarity – monitoring and observability, Zilliz/MilvusDB – vector DB
