@rajeev02/edge-ai

On-device ML toolkit with Indian ID card OCR (Aadhaar/PAN/DL), model pipeline (TFLite/CoreML/ONNX/WASM), and voice AI in 11 Indian languages.

Part of Rajeev SDK — cross-platform infrastructure libraries for building apps that work everywhere.

Why use this?

• Indian ID OCR — Extract and validate Aadhaar, PAN, Driving License from scanned text. Verhoeff checksum for Aadhaar.
• ML model pipeline — Register, load, and run TFLite / CoreML / ONNX / WASM models with a unified API
• Voice AI — Parse voice commands in 11 Indian languages (Hindi, Tamil, Bengali, Telugu, Marathi, Gujarati, Kannada, Malayalam, Odia, Punjabi, English)
• Intent detection — "pay 500 to Rajeev" → { intent: "pay", entities: { amount: "500", recipient: "Rajeev" } }
• GPU acceleration — Optional GPU backend for TFLite and CoreML models
• Privacy-first — All processing happens on device. No data sent to cloud servers.

⚠️ Important: ML Runtimes & OCR Engine Required

This library provides ML pipeline management, text parsing utilities, and voice intent matching. It does NOT include ML models, OCR engines, or speech recognition.

| What the library does | What YOU must provide | | ------------------------------------------------- | --------------------------------------------------- | | Parse OCR text to extract PAN/Aadhaar/DL numbers | OCR engine to convert images → text first | | Validate Aadhaar (Verhoeff checksum), PAN format | Nothing — works standalone | | Model lifecycle management (register/load/unload) | TFLite, CoreML, or ONNX runtime + model files | | Rule-based voice intent parsing (11 languages) | Speech-to-text engine to convert audio → text first |

OCR functions (extractPanNumber, extractAadhaarNumber, detectDocumentType) accept pre-extracted text strings — you must first convert images to text using an OCR engine:

| Platform | Recommended OCR engine | | -------------- | ---------------------------------------------------------------------------------------------- | | iOS | Apple Vision Framework | | Android | Google ML Kit Text Recognition | | Cross-platform | Tesseract.js (web/WASM) |

ML pipeline manages model state (register → load → ready → unload) but actual inference must be implemented via native ML runtimes on each platform.

Voice AI parses text commands using regex-based intent detection — it does not record or transcribe audio. Feed it text from a speech-to-text engine like Google Speech-to-Text or Whisper.

Platform Support

| Platform | Engine | Status | | ---------- | ---------- | ------ | | iOS 16+ | TypeScript | ✅ | | Android 7+ | TypeScript | ✅ |

Installation

npm install @rajeev02/edge-ai

Peer Dependencies

• react >= 18.3.0
• react-native >= 0.84.0 (optional)

Quick Start

Indian ID OCR

import {
  detectDocumentType,
  extractPanNumber,
  extractAadhaarNumber,
  validatePan,
  validateAadhaar,
  getScanningTips,
} from "@rajeev02/edge-ai";

// Detect document type from OCR text
const docType = detectDocumentType("Income Tax Department PAN ABCPE1234F");
// → "pan"

// Extract PAN
const pan = extractPanNumber("Name: Rajeev PAN: ABCPE1234F DOB: 01/01/1990");
// → "ABCPE1234F"

// Extract masked Aadhaar
const aadhaar = extractAadhaarNumber("Aadhaar: 1234 5678 9012");
// → "XXXX XXXX 9012"

// Validate
validatePan("ABCPE1234F"); // → true
validateAadhaar("123456789012"); // → true (Verhoeff checksum)

// Get scanning tips for users
getScanningTips("aadhaar");
// → ["Place the card on a flat surface", "Ensure all four corners are visible", ...]

ML Model Pipeline

import { ModelManager } from "@rajeev02/edge-ai";

const mm = new ModelManager();

// Register a model
mm.register({
  id: "face_detect",
  name: "Face Detection",
  modelPath: "models/face.tflite",
  backend: "tflite",
  useGpu: true,
});

// Load into memory
await mm.load("face_detect");
console.log(mm.isReady("face_detect")); // → true

// Get model info
const info = mm.getModelInfo("face_detect");
// → { id, name, status: "ready", backend, sizeBytes, loadTimeMs }

// Unload when done
mm.unload("face_detect");

Voice AI

import { parseVoiceIntent, SUPPORTED_VOICE_LANGUAGES } from "@rajeev02/edge-ai";

// Parse natural language commands
const intent = parseVoiceIntent("pay 500 to Rajeev", "en");
// → { intent: "pay", confidence: 0.95, entities: { amount: "500", recipient: "Rajeev" }, language: "en" }

const hindiIntent = parseVoiceIntent("Rajeev ko 500 bhejo", "hi");
// → { intent: "pay", confidence: 0.92, entities: { amount: "500", recipient: "Rajeev" }, language: "hi" }

// 11 supported languages
console.log(SUPPORTED_VOICE_LANGUAGES);
// → ["en", "hi", "ta", "bn", "te", "mr", "gu", "kn", "ml", "or", "pa"]

Supported Document Types

| Type | Detection Keywords | Extraction | | --------------- | ------------------------------- | -------------------------- | | Aadhaar | "UIDAI", "Aadhaar" | Number (masked), name, DOB | | PAN | "Income Tax", "PAN" | PAN number, name, DOB | | Driving License | "Driving", "License", "DL" | DL number, name, validity | | Voter ID | "Election Commission" | Voter ID, name | | Passport | "Passport", "Republic of India" | Passport number, name |

API Reference

| Export | Type | Description | | --------------------------- | ---------- | ----------------------------------- | | detectDocumentType() | function | Detect Indian ID type from OCR text | | extractAadhaarNumber() | function | Extract masked Aadhaar number | | extractPanNumber() | function | Extract PAN number | | validatePan() | function | Validate PAN format | | validateAadhaar() | function | Validate Aadhaar (Verhoeff) | | getScanningTips() | function | Get user-facing scanning tips | | ModelManager | class | ML model lifecycle management | | parseVoiceIntent() | function | Parse voice commands to intents | | SUPPORTED_VOICE_LANGUAGES | string[] | List of supported languages |

Full Documentation

📖 Complete API docs with model config and voice AI

License

MIT © 2026 Rajeev Kumar Joshi