01 — Project
Retail AI Orchestrator
Multi-agent retail chatbot that routes plain-English questions to domain agents pulling live .NET ERP data.
100
Daily Chats / User
14s
Max Response Time
16
Max Tool Iterations
24h
Session TTL
40+
.NET Endpoints
10+
Domain Agents
01 — Overview
High level system flow
From user query to intelligent response — each step is optimized for accuracy, speed, and reliability.
User Input
Natural language question
Intent Pre-gate
Clarify metric, date, shop
Router
Rules + LLM classification
Agent Execution
Tool calls to .NET ERP
Data & Synthesis
Densify & analyze results
User Output
Business-ready answer
User Input
Natural language question
Intent Pre-gate
Clarify metric, date, shop
Router
Rules + LLM classification
Agent Execution
Tool calls to .NET ERP
Data & Synthesis
Densify & analyze results
User Output
Business-ready answer
Tech: OpenAI Function Calling · Pydantic v2 · HTTPX · Redis · JWT Auth
02 — Impact
Business value
Empowering retail teams with accurate, real-time answers across all critical operations.
The Problem
Retail ops data spans 10+ domains across a complex .NET ERP — a single LLM prompt hallucinates KPIs and APIs.
The Solution
Multi-agent orchestration with rule + LLM routing, deterministic shop/date resolution, and validated .NET tools.
The Result
Business-ready answers with agent attribution instead of raw API dumps.
Max Response Time
LLM Timeout
API Timeout
Reliability Target
System Availability
03 — Architecture
System architecture
Polyglot hybrid architecture bridging Python FastAPI agents with a .NET ERP (Footzy).
Client / Frontend(Footzy Web)
Web App
JWT + Cookie
Forwarding
API Gateway
FastAPI
Router /
Orchestrator
Domain Agents
(10+ Agents)
Net API Client
(HTTPX)
Redis
Session & Memory
.NET ERP (Footzy)
40+ REST Endpoints
Sales · Traffic · Cameras
Compliance · Hardware
& more
Company Scoped
Secure APIs
Client / Frontend
Footzy Web · JWT + Cookie
API Gateway
FastAPI
Router / Orchestrator
Domain Agents
10+ Agents · Redis session
Net API Client
HTTPX
.NET ERP (Footzy)
40+ REST Endpoints
04 — System Map
Fyen end-to-end architecture
How a store manager's question becomes a business-ready answer — conversation layer, orchestrator, specialist agents, and the .NET ERP ecosystem underneath.
Conversation layer
Natural-language chat with context, security, and sticky sessions.
Router & agents
Intent routing into domain specialists for sales, traffic, products, and more.
.NET ERP fabric
Tool calls into Footzy endpoints, SQL, vectors, Redis, and external services.
Query → Fyen Assistant → Router → domain agents → secure tool calls → Footzy ERP, databases, and memory → structured natural-language response.
05 — Agents
Domain specialists
Each agent has a focused toolset and context for better accuracy.
Sales & Basket
KPIs, ATV, VPSH, Units, IPT
Traffic & Conv.
Footfall, conversion, peak hours
Shop Compare
Rankings, drill-downs, comparisons
Demographics
Age & gender analytics
Products
Category, dept, SKU performance
+5 More
Shifts, Cameras, Uniform, Hardware Insights, etc.
06 — Performance
Performance snapshot
Operational limits and tuning targets from production configuration.
Response Time (Max)
14s
LLM Timeout
60s
API Timeout
30s
Daily Chat Limit
100
Session TTL
24h
Interested in how we built it?
Multi-agent retail chatbot that routes plain-English questions to domain agents pulling live .NET ERP data.