← Back to portfolio

MemeMatch experiment

Karolina

Character with memory and humor, built as an interaction experiment for MemeMatch.

Overview

Karolina was an experiment with a persona that understands conversation context, remembers the user, and answers with more character than a generic chatbot. It first ran on Make, Airtable, and Telegram, then moved toward Python, FastAPI, and Firebase.

Problem

For MemeMatch, the matching mechanic alone was not enough. I wanted to test whether a character with memory, humor, and a clear voice could increase engagement and give the product a stronger personality.

Product bet

If AI remembers the user and has a clear role inside the product, it can become part of the experience instead of another chat box in the corner.

What I built

  • First MVP on Make, Airtable, and Telegram.
  • User memory, forgetting outdated facts, and contextual answers.
  • Python, FastAPI, Firebase Auth, and Firestore version.
  • Voice, support, and social-content experiments.

AI layer

The LLM answered through a specific persona, using user memory and context. The important part was interaction style: less generic assistant, more character that fits MemeMatch.

Architecture

Make, Airtable, Telegram, Python, FastAPI, Firebase Auth, Firestore, OpenAI/GPT, and ElevenLabs experiments.

Key decisions

  • Treat the persona as part of the product, not prompt decoration.
  • Test behavior in a simple MVP first.
  • Avoid selling it as a standalone product without a stronger use case.

Outcome / evidence

Experiment with memory, humor, support, and social content. In this portfolio I treat it as a MemeMatch module, not a separate product.

What I would do next

Return to Karolina when MemeMatch or a similar product needs its own AI character.

Contact

Tell me what you want to build.

Two sentences about the idea are enough. I’ll write back with where I would start, what I would skip, and whether I’m even the right person for it.

Email me