dwani.ai - Discovery
dwani.ai provides secure AI document analytics tailored for proprietary data.
Overview
dwani.ai empowers users with the following features:
- Text and Image Inference: Process and analyze text and images with state-of-the-art models.
- Document Processing: Extract, translate, and query documents efficiently.
- Speech Processing: Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) for Indian + European languages.
- Translation: Seamless translation across major Indian + European languages.
- Scalability: Load balancing and API orchestration for robust performance.
System Architecture
The dwani.ai engine is designed for modularity and scalability. Below is a high-level overview of the inference pipeline:
Component Setup - Discovery
Each component is modular and can be set up independently. Refer to the linked guides for detailed setup instructions.
| Component | Description | Setup Guide |
|---|---|---|
| vLLM Server | Text and image inference | vLLM Deployment |
| API Server | API gateway & Swagger setup | API Server Setup |
| Discovery Server | Discovery Server | Discovery Setup |
| Proxy Server | Load balancer | Proxy Server Setup |
Component Setup - Multimodal Inference
For multimodal inference, the following components are available. Each is independently configurable.
| Component | Description | Setup Guide |
|---|---|---|
| Docs API Server | Document extraction, translation, query | Docs API Setup |
| Translate Server | Indian language translation | Translate Server |
| TTS Server | Text-to-Speech (Indian languages) | TTS Server |
| ASR Server | Automatic Speech Recognition | ASR Server |
Model Dependencies
The following models power Dwani.ai's functionality:
| Task | Models Used |
|---|---|
| Text + Vision | google/gemma-3-27b-it, google/gemma-3-12b-it, google/gemma-3-4b-it |
| Text | Qwen/Qwen3-32B, Qwen/Qwen3-14B |
| Vision | vikhyatk/moondream2 |
| Speech Synthesis (TTS) | ai4bharat/IndicF5, onnx-community/Kokoro-82M-v1.0-ONNX |
| Translation | ai4bharat/IndicTrans3-beta, ai4bharat/indictrans2-indic-indic-1B, etc. |
| Automatic Speech Recognition (ASR) | ai4bharat/indic-conformer-600m-multilingual, Systran/faster-whisper-large-v3 |
| Text Analysis | ai4bharat/Cadence |
Contributing
We welcome contributions! Please open issues or submit pull requests to improve dwani.ai.
License
This project is licensed under the MIT License.
Built with ❤️ by the dwani.ai team