xai-workbench

Project Overview

xai-workbench (formerly "Lingxi" SME AI Workbench) is an open-source, enterprise-grade AI workbench designed to empower businesses of all sizes to leverage the power of Generative AI without requiring deep AI expertise. It acts as a comprehensive SaaS platform that integrates various generative AI capabilities, provides pre-built templates for common business scenarios, and offers robust low-code/no-code customization features.

Our goal is to democratize AI application development, enabling enterprises to deploy, fine-tune, and evolve AI models securely and efficiently within their own environments, thus transforming internal operations and external interactions.

• 中文版

Customer Pain Points & Value Proposition

1. Lack of AI Talent & Localized Models

• Pain Point: Enterprises struggle to build and train AI models tailored to their specific business scenarios due to a shortage of AI professionals and quality data. Reliance on generic large models often yields suboptimal results and higher costs.
• Value: xai-workbench makes "everyone an AI application master" by providing "out-of-the-box" scenario-based AI models and templates. This significantly reduces the barrier to entry and customization costs.
  • Examples: Marketing copy generation, intelligent customer service scripting, product introduction video drafts, simple design element generation, contract document intelligent comparison, code-assisted generation and debugging.

2. Data Security & Privacy Risks

• Pain Point: Uploading sensitive enterprise data to public large models poses significant leakage risks, and there's no guarantee that data won't be retained by third parties.
• Value (Secure & Controllable):
  • The platform offers a secure sandbox environment where all data remains within the enterprise's dedicated space for fine-tuning and microservice calls.
  • Supports private deployment options, allowing models and data to be hosted on the customer's private cloud or servers, completely eliminating data leakage concerns.

3. Insufficient Model Evolution & Customization

• Pain Point: Enterprise needs are dynamic, and a one-time fine-tuned model cannot continuously adapt and optimize.
• Value (Continuous Evolution):
  • Supports online fine-tuning and rolling updates, continuously incorporating user feedback and real business data into training samples.
  • Provides a "My AI Assistant" interface, enabling users to quickly customize exclusive models through natural language descriptions or drag-and-drop workflows.

Key Features

1. Scenario-based Pre-built Templates & Rapid Deployment: Access an "AI App Marketplace" with ready-to-use modules for marketing, customer service, operations, R&D, and more.
2. "My AI Assistant" Deep Customization:
   • Low-code Drag-and-Drop: Visually build exclusive business workflows (e.g., "Text Input → Data Cleaning → Model Fine-tuning → Deployment").
   • Natural Language Fine-tuning: Describe your needs in natural language, and the platform automatically selects the best fine-tuning data and model parameters for online training.
   • Incremental Learning: Continuously optimize exclusive models based on user feedback and new business data.
3. Knowledge Base Augmentation & Enterprise Data Integration:
   • Upload internal knowledge bases (e.g., sales scripts, technical documents, SOPs) for automatic indexing and retrieval.
   • Enhance generation quality with Retrieval-Augmented Generation (RAG), ensuring content relevance to enterprise business.
   • Configurable knowledge base update strategies and data anonymization.
4. Security & Compliance:
   • Data Anonymization & Access Control: Automatic anonymization of sensitive data, with role-based access control.
   • Content Auditing & Risk Filtering: Built-in sensitive information detection to ensure compliance.
   • Private/Hybrid Deployment: Flexible deployment options to keep data within the enterprise network.
5. Multi-role & Permission Management: Supports distinct roles like enterprise administrator, business user, AI operations, and data auditor, each with granular permissions.
6. Visual Monitoring & Metric Feedback: Real-time visualization of model/task usage, latency, quality scores, and user satisfaction. KPI monitoring with daily/weekly reports.

Architecture Overview

For a detailed understanding of the system's architecture, including its layered design, core components, and interactions, please refer to: docs/architecture.md

Building and Running Guide

Detailed instructions for setting up your development environment, building the project, and running various components will be provided here. This section will cover:

1. Prerequisites: Go (1.20.2+), Docker, Kubernetes (optional for full deployment), Git.
2. Local Development: Steps to run individual services.
3. Containerization: How to build Docker images.
4. Kubernetes Deployment: Helm charts or Kustomize configurations for cloud deployment.

(Placeholder for actual instructions in future iterations)

Contributing Guide

We welcome contributions from the community! If you're interested in improving xai-workbench, please refer to our contributing guidelines: CONTRIBUTING.md

(Placeholder for actual instructions in future iterations)

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.