🌌 AstroVoice-Agent

AstroVoice-Agent is a real-time, agentic voice AI middleware that bridges ASR (Speech-to-Text) and TTS (Text-to-Speech) systems, while orchestrating intelligent conversations using LLMs and asynchronous pipelines.

It is designed for telephony-grade voice systems where responses must be fast, stateful, and structured.

---

🧠 System Overview

User Speech
   ↓
ASR (Speech → Text)
   ↓
AstroVoice-Agent (LLM + LangGraph)
   ↓
Async Intelligence Pipeline (Astrology)
   ↓
Response
   ↓
TTS (Text → Speech)

---

⚙️ Core Capabilities

• Agentic orchestration using LangGraph (stateful, conditional routing)
• Voice-first responses optimized for real-time conversations
• Structured data extraction (Name, DOB, TOB, POB)
• Async pipeline execution triggered only when required inputs are complete
• Strict prompt-governed reasoning (no hallucinated astrology)
• State management via Redis (conversation, session, extracted data, computed charts)
• Message-level control using reply_id for deletion

---

🏗️ Architecture

Layers

Layer	Responsibility
API Layer	FastAPI endpoints, request handling
Agent Layer	LangGraph workflow, LLM execution
State Layer	Redis (sessions, memory, charts)
Pipeline Layer	Astrology computation + profile generation
LLM Layer	Azure OpenAI / OpenAI-compatible models

---

📁 Project Structure

astrovoice-agent/
│
├── config.py      # Redis, API keys, constants
├── models.py      # Pydantic models + TypedDict state
├── astro.py       # Geocoding, Prokerala APIs, profile builder
├── graph.py       # LangGraph nodes, routing, compiled graph
├── main.py        # FastAPI app, endpoints, session handling
│
├── requirements.txt
└── README.md

---

🔄 Agent Workflow

        ┌────────────┐
        │  LLM Call  │
        └─────┬──────┘
              │
     ┌────────▼────────┐
     │   Route Logic   │
     └─────┬─────┬────┘
           │     │
           │     ▼
           │  Summarize → END
           │
           ▼
   Astro Pipeline (async)
           │
           ▼
          END

---

🚀 Setup

1. Clone repository

git clone https://github.com/ut2903/AstroVoice-Agent.git
cd AstroVoice-Agent

---

2. Create virtual environment

python -m venv venv
source venv/bin/activate   # Mac/Linux
venv\Scripts\activate      # Windows

---

3. Install dependencies

pip install -r requirements.txt

---

5. Run application

python main.py

Server runs on:

http://localhost:5000

---

📡 API

POST /chat

Request

{
  "user_id": "123",
  "name": "User",
  "message": "Hello"
}

Response

{
  "Agent": "नमस्ते...",
  "call_status": "ONGOING",
  "language": "Hindi",
  "summary": "",
  "reply_id": "uuid"
}

---

POST /delete_message

Request

{
  "reply_id": "uuid"
}

Behavior

• Deletes the AI response
• Deletes preceding user message (if applicable)
• Maintains conversation consistency

---

🧠 Design Constraints

• Astrology interpretation is strictly based on pipeline-generated data
• No inference from DOB/TOB/POB without computed chart
• Responses are short and voice-friendly
• JSON output format is enforced for every turn
• Pipeline executes only once per user (cached in Redis)

---

⚡ Tech Stack

• FastAPI
• LangGraph
• Azure OpenAI / OpenAI-compatible LLM
• Redis
• External APIs (Geocoding + Astrology)

---