Browser‑Based AI Models for Production Use

Описание и применение

Модели ИИ, работающие в браузере, можно использовать в продакшене для задач, таких как удаление фона, генерация альтернативного текста, создание структурированных тикетов и семантический поиск. Они запускаются локально через WebGPU/WebAssembly, не требуют внешних API и сохраняют конфиденциальность данных.

Задача

Can small AI models run directly in the browser be useful for real production applications?

Результат

Browser‑based AI can deliver practical, privacy‑preserving features in production, especially for e‑commerce, accessibility, and support workflows, though large models remain impractical and licensing must be checked.

Рабочий процесс

• 1. Set up a React + TypeScript app with Web Workers for inference.
• 2. Load models on demand via Transformers.js (e.g., RMBG‑1.4, Florence‑2‑base‑ft, Qwen2.5‑0.5B‑Instruct, all‑MiniLM‑L6‑v2).
• 3. Execute inference using WebGPU where available, fallback to WebAssembly CPU.
• 4. Provide UI for each feature (e.g., upload image → background removal, generate alt text, describe issue → structured ticket, search tickets).
• 5. Validate structured outputs with schema validation (Zod).
• 6. Allow user review/edit before final use.

Доказательства

I wanted to focus on practical applications. More specifically, I wanted to find out how we could use browser-based AI today, right now, in real production applications.

Источник

The models are loaded on demand, so you don't need to download all five of them just to open the application.

Источник

RMBG‑1.4, with approximately 44 million parameters, does a really impressive job of separating an object from its background.

Источник

Florence‑2‑base‑ft, a vision model with approximately 230 million parameters, to generate descriptions of images.

Источник

Qwen2.5‑0.5B‑Instruct, a small language model with approximately 500 million parameters.

Источник

all‑MiniLM‑L6‑v2, a small embedding model with approximately 22 million parameters.

Источник

The application searches for tickets with a similar meaning. And, of course, I also included a traditional keyword search so you can compare the results yourself.

Источник