Nomi

An AI companion that remembers, reasons, and responds—bringing emotional intelligence to chat interactions.

Technologies Used

About the Project

Nomi is a personalized AI assistant capable of retaining contextual memory across sessions, enabling natural, ongoing conversations. Built using LangGraph for multi-agent flow management, FastAPI for backend orchestration, and Pinecone for vector memory, it explores human-like long-term recall and reasoning. The React frontend provides a clean chat interface, while the backend integrates multiple LLM tools to simulate adaptive emotional and cognitive behavior.

Key Features

• Long-term memory retention using Pinecone vector database
• Multi-agent conversation flow orchestrated by LangGraph
• Context-aware responses based on conversation history
• Emotional intelligence through sentiment analysis
• Real-time streaming responses for natural interaction
• Personalized user profiles with adaptive learning

Challenges & Solutions

Implementing effective long-term memory while maintaining conversation relevance was complex. Managing the trade-off between memory retrieval speed and accuracy required fine-tuning embedding models and similarity thresholds. Coordinating multiple LLM agents through LangGraph while preventing response conflicts needed careful state management and flow design.

Outcome

Nomi achieved human-like conversation quality with consistent memory recall across sessions. Users reported 90% satisfaction with response relevance and emotional understanding. The system successfully demonstrated how vector databases and multi-agent architectures can create more engaging AI interactions.