MentAI
A personalized mentorship app: daily guidance, reflection prompts, and adaptive tasks for cognitive and productivity goals. Seven of us built it — close friends, mostly engineers — over ten months of evenings and weekends. It was my first end-to-end product. We stopped it after MVP testing.
What it did
The idea worth keeping was structured user state instead of chat history: a user wasn't a transcript but a profile, goals, daily inputs, assigned tasks, and a weekly summary acting as compressed memory. Serverless functions on Google Cloud handled the asynchronous work — processing a day's input, scheduled nudges, weekly reviews — and the LLM sat inside controlled pipelines as a reasoning component. The architecture worked. That is not the same as the product working.
Why we stopped
Around twenty friends tried it on TestFlight; it never reached the App Store. The daily letters — the core feature — became annoying, and I wasn't reading them either, on the product I had built for my own problem. Interviews stayed encouraging while usage said otherwise: people tried it, drifted, and didn't miss it.
We were generating mentorship content and none of us knew anything about mentorship. The users we could reach didn't have the daily need the whole design assumed, and no iteration on our side could manufacture it. At some point we said it out loud: none of us needed this product. That was the end of it.
One thing did work: we grew the MentAI Instagram to around four thousand followers organically — which taught us that distribution is a separate craft from building, and why ConsulDent went to cohorts of real dentists instead of an audience.