Inspiration
Most of us are quietly not okay sometimes, and we rarely do anything about it in the moment. There is a wide gap between feeling stressed on an ordinary Tuesday and actually booking a therapist. We wanted to build something small and kind that lives in that gap. Not a replacement for real care, just a gentle companion that helps you notice a feeling and do one small thing about it.
How we built it
We made it a Telegram bot, so there is nothing new to install and no extra habit to remember. You already open Telegram, so caring for yourself can happen there too. You log how you feel, you get a short coping exercise, you move through it one small step at a time, then you rate the feeling again. Over time the bot learns which exercises actually help you for each emotion and quietly moves those to the top. The whole core loop runs with no AI at all, so it stays dependable, and we layered Gemini on Vertex AI on top for the human parts: naming a feeling from something you typed, pulling exercises out of a PDF you already trust, and writing a warm weekly reflection. Everything is stored privately per person, and we never log what you write.
What we learned
Building the core loop to work with no AI first was the best decision we made. It kept the bot dependable and turned the AI into something optional rather than load bearing. Making every AI call fail softly, and parsing model output into typed data instead of raw text, saved us from a lot of fragile behaviour. Clear module boundaries let two of us build in parallel without breaking each other's work. We also found that keeping all the user facing wording in one place paid off, since small phrasing choices genuinely change how supportive the bot feels.
The hardest part
Staying honest about what this is and is not. It is easy to drift toward sounding like a therapist, and we never wanted that. Keeping the AI optional, making it fail softly, and making the bot feel patient rather than like a form all took real care. With two people and very little time, clean boundaries are what kept it calm.
What's next for Everyday-Im-Okay
We want the bot to learn faster and ask less of you. That means adaptive recommendations that balance what has already worked for you with trying something new, and that learn across related emotions. It means short coping exercises the AI writes on the spot, shaped by the moment and your history. We also want optional lightweight context so the bot can surface the triggers behind a feeling, like work, sleep, or people. And we want voice journaling, so in a hard moment you can simply send a voice note and let the bot help you name the feeling. The bigger idea is a bot that reaches out before you struggle, gently. With your calendar connected, it could link recurring feelings to the events behind them and offer a reset before the kind of meeting that usually spikes your stress. With health data such as heart rate, it could notice the tension your body shows before your mind names it and suggest the right exercise in the moment. We also want a positive habit tracker that runs alongside the journal, helping good practices grow until they crowd out the bad ones. Further out, the modular core makes it cheap to add more languages and reach beyond Telegram, and we want a user controlled summary you can export and bring to a therapist, so the bot becomes a bridge to real care rather than a wall in front of it.
Log in or sign up for Devpost to join the conversation.