Perception AI

Perception AI is a production-ready, full-stack conversational AI platform. It features a modern React frontend, a robust FastAPI backend, and advanced agentic capabilities powered by LangGraph.

🌟 Key Features

• Advanced Chat Interface:
  • Real-time streaming responses (SSE).
  • Markdown support, code highlighting, and typing indicators.
  • Persistent chat history and session management.
• Voice Mode 🎙️:
  • Hands-free voice interaction.
  • Real-time Speech-to-Text (Whisper) and Text-to-Speech (OpenAI/ElevenLabs).
  • Modern, animated voice overlay.
• Agentic Intelligence:
  • Powered by LangGraph for complex reasoning.
  • Integrated tools for web search and calculation.
  • Context-aware conversations with long-term memory.
• Secure & Scalable:
  • JWT Authentication (Access/Refresh tokens).
  • Redis caching and rate limiting.
  • PostgreSQL database with async support.
• Branch-Your-LLM 🌳:
  • Non-linear Conversations: Fork chats at any point to explore alternative paths.
  • Tree Visualization: Interactive graph view of your entire conversation history.
  • Context Management: Each branch maintains its own independent context.
  • Regeneration: Get multiple AI responses for the same prompt and compare them.

🌳 Branch-Your-LLM Guide

The Branch-Your-LLM system transforms linear chats into powerful conversation trees.

Key Features

• Fork: Create a new branch from any message to explore a different direction.
• Tree View: Switch to the "Tree" tab to see the full conversation graph.
• Navigate: Click any node in the tree to jump to that point in the conversation.
• Regenerate: Create sibling nodes to compare different AI responses.

How to Use

1. Start a Chat: Begin a conversation as usual.
2. Open Tree View: Click the GitBranch icon (🌿) in the top-right header.
3. Sync: If asked, click "Sync Chat to Tree" to visualize your current conversation.
4. Explore: Use the interactive tree to navigate different conversation paths.

Note: Some advanced branching features are currently in beta.

For setup instructions, see SETUP.md.

🔬 Deep Research Mode

Deep Research Mode is an advanced AI-powered research feature that performs iterative, evidence-backed research.

Key Features

• Iterative Research: Performs multiple rounds of research to deepen understanding.
• Evidence-Backed: Extracts and verifies claims against real-world sources (Tavily).
• Gap Analysis: Automatically identifies missing information and generates new queries.
• Structured Reports: Generates comprehensive JSON reports with executive summaries, findings, and references.

How to Use

1. Navigate: Go to /deep-research in the application.
2. Configure:
   • Topic: Enter your research question.
   • Depth (1-5): Set the level of detail.
   • Iterations (1-10): Set the thoroughness.
3. Start: Click "Start Research" and watch the real-time progress.

📊 Real-Time Progress Tracking

We've implemented real-time progress tracking for both Deep Research mode and regular chat, providing users with transparent visibility into what the AI is doing at every step.

🔬 Deep Research Tracking

Shows detailed, step-by-step progress during deep research iterations:

• Searching 🔍: Web search with query display and source counts.
• Extracting 📄: Claim extraction from sources.
• Verifying ✅: Evidence-based claim verification.
• Analyzing 🧠: Knowledge gap analysis.
• Synthesizing ✨: Final report generation.

💬 Chat Progress Tracking

Shows real-time agent actions during regular chat conversations:

• Thinking 🧠: Processing your request.
• Searching 🔍: Web search with actual queries.
• Reading 📄: Analyzing sources (shows domains).
• Analyzing 🧠: Processing information.
• Completed ✅: Task finished.

The UI features smooth animations, color-coded steps, and detailed data display (source counts, actual queries, etc.) to build trust and engagement.

📚 Documentation

We have organized the documentation into the following guides:

• SETUP.md: Installation, configuration, and getting started guide.
• ARCHITECTURE.md: Detailed system architecture, diagrams, and data flow.

🛠️ Tech Stack

Frontend (/client)

• Framework: Next.js / React
• Styling: Tailwind CSS, Shadcn UI
• State: Zustand
• API: Axios, EventSource

Backend (/server)

• Framework: FastAPI
• Database: PostgreSQL (SQLModel), Redis
• AI: LangGraph, Groq (Llama 3), OpenAI (Whisper/TTS)
• Auth: Python-JOSE (JWT), Passlib (Bcrypt)

🚀 Quick Start

# 1. Start Backend
cd server
python main.py

# 2. Start Frontend
cd client
npm run dev

For detailed instructions, see SETUP.md.

📄 License

MIT License