Hey everyone,
I’m working at Suffescom Solutions, and I’ve been developing a Candy AI Clone with custom AI characters and conversation flows. Overall, the project is progressing well, but I’ve hit a confusing issue.
The AI characters fail to maintain a consistent personality. After a few messages—or when the chat session reloads—the tone, behavior, and responses start drifting, even though each character has a fixed personality prompt.
Tech stack:
-
Backend: Python + FastAPI
-
AI: OpenAI API
-
Frontend: React
What I’ve tried so far:
-
Fixed system prompts per character
-
Storing personality traits in the database
-
Limiting context window size
Despite this, the personality inconsistency keeps happening.
Has anyone faced this while building a Candy AI clone or character-based chatbot?
Would really appreciate advice or best practices.