MimiBot
MimiBot was a Telegram AI assistant I built roughly from September 2025 through January 2026. The aim was practical and personal: reduce the friction between noticing a study problem and doing something useful about it.
The experiment combined Telegram integration with conversational AI and persistent, personalized memory experiments. It could work with image input as well as text, which made it possible to try study workflows that did not begin with a cleanly typed prompt. The project also included study tools and a math graphing workflow that could start from either an image or a written expression.
Direct Telegram access was useful for this personal experiment because it kept the assistant close to the moment a study question appeared. Image input offered a similar path: when a question or expression was already in an image, I could explore moving from that image to a graph without retyping it. These choices were experiments, not claims about production performance or a user base.
What I was exploring
MimiBot was less a finished product than a way to test how an assistant might preserve useful context, adapt to an individual's study habits, and turn everyday questions into small actions. Memory was deliberately an area of experimentation: the point was to learn what persistent context helped, where personalization became noisy, and how different inputs changed the interaction.
Status: experimental personal study system. It should be understood as a private learning project and not as a production platform or a general purpose Telegram service.