Welcome to Pravah
Pravah is a visual multi-node Indic speech and language pipeline studio. Powered by Sarvam AI APIs and backed by high-performance Neon PostgreSQL storage, Pravah enables developers, product managers, and researchers to design, chain, execute, and audit speech recognition, machine translation, script transliteration, and voice synthesis workflows.
AI Pipeline Builder
Explain your desired translation or speech workflow in natural language (voice or text), and Pravah AI will instantly wire the nodes together.
Real-time SSE Monitor
Watch execution tokens flow from node to node live in the visual canvas. Spot processing bottlenecks or failures instantly.
Onboarding Questionnaire
Upon signing up, Pravah walks you through a 3-step onboarding process to customize your workspace dashboard. This information helps seed relevant templates and allocate trial credits.
Profile & Role
Select your functional background (AI/Software Engineer, Product Manager, Researcher, Enterprise Architect). This determines your startup template defaults.
Use Cases & Goals
Pick your specific pipeline objectives (Voice Assistants, Audio/Document Translation, Text-to-Speech narration, Custom AI logic).
Languages & Scale
Select primary target Indic languages (e.g. Hindi, Telugu, Tamil) and indicate expected operational scale (Starter, Growth, Enterprise).
Managing Projects
Projects act as isolated sandbox workspaces containing related pipelines, local R2 assets, and transaction audits.
To Create a Project:
- Navigate to the Projects tab from the main navigation header.
- Click the New Project button on the top right.
- Provide a descriptive Project Name and optional description.
- Click Create Workspace. Your project card will appear in the grid.
Every project is automatically seeded with custom quick-access templates from the Library Project if available, allowing you to instantly boot up common patterns like:
- đĨ Patient Voice Translation (Hinglish/Tenglish to English)
- đ° Multilingual Document OCR & Summary
- đī¸ Indic Conversation Podcast Generator
Creating Pipelines
Pravah offers two distinct ways to create data pipelines inside your project workspaces: **AI-Powered Generation** and **Manual Canvas Building**.
Option A: Pravah AI Pipeline Builder
Generate an entire connected pipeline using natural language. The AI understands speech contexts and Indic translation schemas, arranging nodes dynamically.
AI Builder Features:
- Voice Prompts: Click the microphone icon to record your pipeline requirements directly. Supports speech recognition in English, Hindi, Telugu, Tamil, and Malayalam.
- Audio Upload: Upload an audio file (.wav, .mp3) outlining what you want to build.
- Visual Preview: View a schematic layout showing the generated nodes and connection edges before saving the pipeline to your project.
"Create a pipeline that takes a public URL, scrapes the content, summarizes it using an LLM in Hindi, and then synthesizes the Hindi summary to speech so I can listen to it."
Option B: Manual Construction
Set up pipelines manually by inputting a pipeline name, choosing a template (or choosing "Blank Pipeline"), and saving. You are immediately directed to the visual layout workspace where you can build step-by-step.
Grid Navigation & Node Placement
The visual editor workspace is built using @xyflow/react. Interactions are highly fluid and responsive:
Toolbar Select
Click any node type from the bottom dock toolbar (Inputs, Processing, Outputs, Integrations) to drop it onto the grid canvas.
Drag Canvas
Left-click and hold a node's title handle, then drag to reposition it. Canvas updates its coordinates in real time.
Node Handles
Drag from the circular **source handle** (right side of node) and drop onto a target **input handle** (left side) of a downstream node.
Deleting Elements
To delete a node or connection line, click to select it (it will show a highlighted border). Press the **Backspace** or **Delete** key on your keyboard, or click the **Delete Icon** in the node's settings panel.
Visual representation of connections:
Configuration Panel
Clicking any node open its settings in the **Configuration Panel** on the right side of the visual editor screen. Here you can tweak specific model parameters and load test values.
Tweak Model Settings
Configure target speaker voices (e.g. Aditya, Ritu, Pooja) for speech synthesis, select translation target/source language codes (e.g. Assamese, Gujarati), or define prompt weights for LLM models.
Dynamic Prompt Interpolation
Downstream reasoning models (LLM) can dynamically inject text outputs from upstream nodes using brackets notation.
Translate the following text into English:
{{node_stt_584.output}}
Interactive Sound Check
In inputs and Text-to-Speech configurations, you can record trial voice prompts using your browser microphone directly within the panel, or play sample voice syntheses to review speaker speeds.
Execution Status Monitor
Running a pipeline starts an SSE (Server-Sent Events) network stream that executes nodes concurrently based on dependency mapping.
Real-time Node Glow states:
Aborting Runs: While running, you can click the **Cancel** button on the bottom of the logs sidebar. This terminates the network stream, sends an abort request to active processes, and registers a warning log.
Credits & Billing Model
Pravah operates on a pay-as-you-go developer model. The visual studio calculates cost estimations before each execution by inspecting node types.
| Node Type | Cost per Execution | Service Model |
|---|---|---|
| Speech-to-Text (STT) | âš0.375 | Saaras:v3 AI |
| Text-to-Speech (TTS) | âš0.050 | Bulbul:v3 AI |
| Translate | âš0.050 | Indic Translation |
| LLM & OCR Nodes | âš0.500 | Generative Hub |
| Inputs & Outputs | Free (âš0.00) | Local Assets / R2 |
Audit Trails & Execution History
Every pipeline run logs execution metrics to the Neon PostgreSQL database, ensuring secure audit tracking.
Accessing logs:
- Click the **History** button in the visual editor toolbar header.
- You will see the **Pipeline Execution History** table.
- Click any run row to expand and inspect the run duration and exact node inputs/outputs.
{
"runId": "run_9f27d4c8",
"status": "COMPLETED",
"startedAt": "2026-08-17T17:42:00.000Z",
"finishedAt": "2026-08-17T17:42:08.500Z",
"nodeRuns": [
{
"nodeId": "node_stt_584",
"status": "COMPLETED",
"output": {
"text": "ā¤šāĨā¤˛āĨ, ā¤ā¤Ē ā¤āĨā¤¸āĨ ā¤šāĨā¤?"
}
}
]
}Update Profile Details
Configure your personal developer identity and track invoices under the **Profile Details** page.
Profile Controls:
- Name Update: Click the edit name inputs and type your name, then save.
- Billing History: View Razorpay payment transaction timestamps.
- Wallet Recharges: Add top-up values (âš100, âš500, âš1000) using the Razorpay gateway modal.