DIGITAL LEARNING

DIGITAL MARKETING

Software DEVELOPMENT

www.8wave.it

WE ARE HIRING!

Client Side AI

Smaller, optimized models for better applications.

Safety & Privacy

Data stays local, no third-party access.

Cost Reduction

No cloud, no bills to pay.

POWER CONSUMPTION

Great results with a fraction of energy.

SOURCE:
https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/

1 mail of 100 word written with GPT-4: 0,14kWh (14 LED lightbulbs for 1h)

> npm install @xenova/transformers

> npm install @tensorflow/tfjs

WEB & CLIENT SIDE AI

What web can do today.

```
speech recognition
```
```
audio classification
```
```
text to speech
```

Natural language PROCESSING

COMPUTER VISION

AUDIO

```
text classification
```
```
question answering
```
```
summarization
```
```
translation
```
```
text generation
```
```
text toxicity
```

```
image classification
```
```
object detection
```
```
image segmentation
```
```
face detection
```
```
hand detection
```
```
depth estimation
```

> npm install @volverjs/ai

@volverjs/ai

npm.im/@volverjs/ai

VolVER family

Maintained with ❤️ by 8 Wave

OPEN SOURCE

It's

BABY

import { Translator } from '@volverjs/ai'

const text = 'Ciao, come stai?'

// init web worker
const translator = new Translator()

// wait for the translation
const result = await translator.translate(text, {
    from: 'ita_Latn',
    to: 'eng_Latn'
})

Translator

import { RemoveBackground } from '@volverjs/ai'

const sourceImageURL = 'https://example.com/image.jpg'

// init web worker
const removeBackground = new RemoveBackground()

// wait for image blob prediction
const resultImageBlob = await removeBackground.predict(sourceImageURL)

// get image URL from blob
const resultImageURL = URL.createObjectURL(resultImageBlob)

REMOVE

BACKGROUND

DEMO

volverjs.github.io/ai

PROBLEMS

WEIGHT

Xenova/nllb-200-distilled-600M

is ~900MB.

SPEED / PoWER CONSUMPTION

Not production ready for mobile devices. Can't access to specific HW (ex. NPU).

REPLICATION

Each website have to download his models.

FUTURE

AI in Web APIs

Google I/O 2024 launched the Early Preview Program for AI features built into Chrome

const session = await ai.assistant.create()

// prompt the model and wait for the result
const result = await session.prompt("Write me a poem")

PROMPT

A full LLM by Gemini Nano.

const summarizer = await ai.summarizer.create()

const someLongText = "The gentle thud reverberated..."

// wait for the result
const result = await summarizer.summarize(someLongText)

SUMMARIZATION

Only in English (by now).

const detector = await translation.createDetector()

const someText = 'Hallo und herzlich willkommen!';
const results = await detector.detect(someText);

for (const result of results) {
  // show the full list of potential languages
  console.log(result.detectedLanguage, result.confidence);
}

Language Detection

Detect +100 languages.

const writer = await ai.writer.create()

const result = await writer.write(
  "A draft of an email to my boss to request a salary increase."
)

Writer and Rewriter

Tools for revising and restructuring text.