The Challenge
A major distributor needed to automate complex business workflows including quotation generation,
inventory queries, and price lookups. The system needed to handle text, voice, and image inputs
through messaging platforms while integrating with existing ERP systems.
Technical Implementation
The full application and AI orchestration is implemented in Rust for speed and efficiency. The system uses
Kimi-K2-Instruct for query understanding with Claude-Sonnet-4 fallback, custom tool functions for
business logic, and real-time WebSocket connections for ERP integration.
<2s
Response Time
99.9%
Uptime
3
Input Modalities
24/7
Autonomous Operation
Key Features
- Multi-modal input processing (text, voice, handwritten images)
- Real-time ERP integration via WebSocket
- Automatic PDF generation with dynamic pricing
- Cost tracking per API call with database logging
- Fallback LLM strategy (Claude → Groq)
- Session-based context management
- Rate limiting and error recovery
- AI workflow observability and debuggability
Tech Stack
Backend: Rust (Tokio, Axum)
AI/ML: Claude API, Groq, AWS Textract, Whisper
Infrastructure: PostgreSQL, WebSocket
Deployment: Docker, DigitalOcean App Platform