CurioCity: A Case Study in Multi-Agent Conversational AI
The Problem
Have you ever been deep into a technical podcast when a critical question pops into your head? By the time you pause to search for an answer, the context is gone and the moment is lost. I built this project to solve that exact problem: to create a learning experience where you can challenge, question, and dive deeper the instant curiosity strikes—in the explanation style that works best for you.
My Solution
1. Real-Time Inquiry
This feature fundamentally changes the user's role from a passive listener to an active participant. The Real-Time Inquiry engine puts you in the director's chair, with the power to interject and steer the conversation towards what's most important to you. It provides targeted, on-demand clarifications without ever losing the main narrative thread.
2. Adaptive Personas
Understanding isn't one-size-fits-all. That's why CurioCity features swappable AI personas—from a direct factual expert to an analogy-driven storyteller. This allows the user to re-frame complex topics on demand, choosing the explanation style that makes the information truly click.
3. Structured Dialogue
CurioCity replaces the overwhelming 'data dump' with a Structured Dialogue. This turn-by-turn flow is deliberately paced to prevent cognitive overload while giving the user significant narrative control. By selecting key themes, users can guide this conversation, ensuring it explores relevant topics in a logical progression.
Tech Stack
Challenges & Learnings
1. Solving Prompt Leakage
[Placeholder]
2. Overcoming Hardware Limitations with GGUF
[Placeholder]
3. Ensuring Factual Grounding with RAG
[Placeholder]
4. Architecting for Stateful Sessions
[Placeholder]
5. Designing for Graceful Conversational Flow
[Placeholder]