Back to Project List

DescartAI

WatonGroup
N/A
Tech Lead / Co-Product Planning
99.95% production availability 38% efficiency gain for account managers; customer service response time cut to under 30 seconds

Project Overview

Aims to provide users with intelligent personal assistant services.

Technical Architecture

  • Backend frameworks: FastAPI. Axum, Gin
  • Programming languages: primarily Python & Golang , with Rust as a secondary language
  • Mobile: Swift (iOS), Kotlin (Android)
  • Web: JavaScript
  • Databases: MySQL, Redis, Milvus
  • AI models: GPT, Claude, Mistral, Grok, Kimi, Qwen, DeepSeek, ChatGLM

Key Features

1. Multi-Model Integration

The project integrates a variety of large language models, including:

  • Domestic (Chinese) models: Qwen, Zhipu, DeepSeek
  • International models: Claude, the GPT family
  • Multimodal models: MiniCPM-Llama3-V and more

2. Data Encryption and Privacy Protection

The project incorporates a complete data encryption and privacy protection mechanism:

  • Phase 1: fixed-key encryption
  • Later phases: more advanced encryption schemes

3. Multimedia Support

The project supports processing of multiple media formats:

  • Text chat
  • Speech to text
  • Text to speech
  • Image recognition and processing

4. Messaging System

A complete message push and handling system is in place:

  • JPush (Aurora Push) integration
  • MQTT service design
  • Message storage and management

Delivery Results

  • Supported 180,000+ multimodal requests in the first month after launch, with a 92% user satisfaction rate.
  • Cut the financial news response pipeline by 65%, letting wealth advisors get compliant talking points within 30 seconds.
  • Passed the annual compliance review of a listed company, including penetration testing and cross-border data audits.

Team and Governance

  • Established a Scrum + OKR collaboration mechanism across five teams: product, AI, mobile, backend, and data.
  • Introduced model release approval, prompt version control, and audit-trail mechanisms to keep production safe.
  • Formed an on-call operations and incident response team, enabling 24/7 monitoring and SLO retrospectives.

Project Progress

Q1 (Jan-Mar)

  • Completed core feature development
  • Integrated RAG technology, using the Millvus vector database
  • Implemented the data encryption scheme
  • Completed full project deployment

Q2 (Apr-Jun)

  • Applied for developer accounts (Apple/Google)
  • Refined the digital human solution
  • Integrated TH (Tonghuashun) features

Q3 (Jul-Sep)

  • Deepened the TH integration
  • Added voice reply
  • Redesigned the API data structures

Q4 (December)

  • Refined the business / data / application / technical architecture
  • Built the message center, decision center, and more

Usage Instructions

Deployment Configuration

# 示例配置
models:
  - name: "qwen"
    type: "domestic"
  - name: "claude"
    type: "foreign"
  - name: "minicpm"
    type: "multimodal"

encryption:
  version: 2
  method: "advanced"

Core Modules

  1. ai-service: core AI service module
  2. ai-secretary: AI secretary module
  3. ai-image-generator: image generation module
  4. message-service: messaging service module