BMS_Documentation_For_Word.md

BMS AI Assistant: System Documentation

1. Executive Summary

The BMS AI Assistant is an enterprise-grade conversational interface designed to optimize supply chain operations. By leveraging local AI inference and deterministic forecasting models, the system provides instant access to critical business data—including demand forecasts, inventory levels, and supplier information—without relying on external cloud APIs.

2. System Architecture

2.1 High-Level Design

The system operates on a micro-service architecture containerized within Docker.

• Presentation Layer: HTML5/JavaScript frontend served via FastAPI.
• Application Layer: Python-based backend handling intent parsing and business logic.
• Data Layer: In-memory CSV data structures for high-performance retrieval.
• Intelligence Layer: Hybrid engine combining ARIMA (Statistical) and TinyLlama (Generative AI).

2.2 Process Flow

1. Input: User submits natural language query.
2. Parsing: Regex engine extracts Intent (e.g., "Forecast") and Entities (e.g., "BMS0015").
3. Routing: Request is directed to the appropriate module (Forecasting, Inventory, etc.).
4. Execution: Module performs computation or data lookup.
5. Response: Results are formatted into JSON and rendered by the frontend.

3. Technical Specifications

Component	Technology	Specification
Backend	Python / FastAPI	v0.104+, Async I/O
Server	Uvicorn	ASGI Standard
Forecasting	Statsmodels	ARIMA(1,1,1)
LLM	TinyLlama	1.1B Parameters, Quantized
Frontend	HTML/JS	Zero-dependency, Vanilla JS
Deployment	Docker	Debian-based Python Slim Image

4. Key Capabilities

Demand Forecasting

• Methodology: Time-series analysis using historical sales data.
• Output: 30-day forward-looking demand schedule.
• Accuracy: Optimized via ARIMA parameter tuning.

Inventory Management

• Scope: Multi-location stock tracking.
• Features: Real-time "On Hand" vs "Allocated" visibility.

Reporting

• Format: PDF (Generated on-demand).
• Content: Chat transcripts, forecast tables, and system alerts.

---

Generated for BMS Project Team | Version 1.0