Agni

README excerpt

# 🎙️ Hinglish Audio Processor

An AI-powered web application that processes spoken Hinglish (Hindi-English code-switched) audio and provides:

- **Accurate Code-Switched Transcript** - Preserves natural Hindi-English mixing
- **Hindi Localization** - Converts to pure Hindi (Devanagari) while preserving meaning, slang, and tone
- **Emotion Detection** - Identifies emotions per segment (excited, frustrated, informative, etc.)
- **Cultural Explanations** - Explains slang, idioms, and cultural terms for non-native speakers
- **Live Playback Sync** - Highlights transcript segments in real-time as audio plays

![Demo Screenshot](demo-screenshot.png)

---

## 🧠 What Makes This Hard

> **The AI processes raw audio directly** — not text. It simultaneously:
> - Transcribes code-switched speech (Hindi + English mixed naturally)
> - Detects emotional tone from voice and word choice
> - Localizes meaning while preserving cultural nuance
> - Identifies and explains slang/cultural references

This is significantly harder than processing pre-existing text because the model must handle:
- Accent variations
- Code-switching boundaries
- Emotional prosody
- Cultural context inference

---

## 🚀 Quick Start

### Prerequisites

- Python 3.10+
- Node.js (optional, for frontend server)
- Google Cloud account with Vertex AI enabled
- Service account with permissions

### 1. Clone & Setup Backend

```bash
# Navigate to backend
cd backend

# Create virtual environment (recommended)
python -m venv venv