Intelligent Multi-Tenant ERP/CRM SaaS Platform with AI
- Date : 2026
- ERP
- Completed
Multi-tenant ERP/CRM SaaS in Spring Boot microservices with an AI service for forecasting, risk scoring, anomaly detection and a local RAG assistant. Team of 4.
Overview
Annual engineering project at IIT Sfax (2026), built by a team of 4: a microservices SaaS ERP/CRM for aluminium joinery companies. It covers CRM, the full sales cycle (quote → order → delivery → invoice → payment), purchasing and inventory.
Each company is a tenant whose data stays strictly isolated, with single sign-on, role-based access control and a tamper-evident audit log. A dedicated AI microservice adds sales forecasting, risk classification, stock anomaly detection and a local RAG assistant.
The platform was delivered in 4 incremental XP releases (120 story points), with CI/CD on GitHub Actions and monitoring via Prometheus and Grafana.
Key figures
- layers of multi-tenant data isolation
- 5
- layers of multi-tenant data isolation
- incremental XP releases
- 4
- incremental XP releases
- story points delivered
- 120
- story points delivered
- team members
- 4
- team members
Technologies
Java
Spring Boot
Angular
PostgreSQL
Keycloak
RabbitMQ
FastAPI
XGBoost
Prophet
Scikit-learn
Ollama
Mistral
RAG
pgvector
Docker
GitHub Actions
Testcontainers
Prometheus
Grafana
Flutter
Valkey
ZipkinProblem
Aluminium joinery companies often manage customers, quotes, orders, stock and purchasing with scattered tools, without a shared view of their activity or help in anticipating sales, payment risks and stock needs.
Solution
Java 21 / Spring Boot 3 microservices behind a gateway, secured by Keycloak (OIDC/PKCE), with an Angular 17 web app; PostgreSQL with Row-Level Security and pgvector; RabbitMQ for event-driven stock synchronization; a FastAPI AI microservice (Prophet, XGBoost, Isolation Forest, Ollama with Mistral 7B); Docker, GitHub Actions CI/CD, Prometheus and Grafana monitoring.
My contribution
Team project (4 people) run with XP: requirements and user stories, microservices architecture, multi-tenant security, business modules, AI microservice, integration tests and CI/CD.
Results
Delivered in 4 incremental XP releases (120 story points). Tenant isolation, SSO and the audit log are validated by Testcontainers integration tests.
Features
- CRM, full sales cycle (quote → order → delivery → invoice → payment), purchasing and inventory
- Multi-tenant data isolation across 5 layers, including a gateway JWT filter, Hibernate filters and PostgreSQL Row-Level Security
- Keycloak SSO (OIDC/PKCE), role-based access control and a SHA-256 hash-chained audit log
- Event-driven stock synchronization via RabbitMQ, business state machines and design patterns (Command, Memento, Strategy, Observer)
- AI microservice: Prophet sales forecasting, XGBoost quote-acceptance and payment-risk classification, Isolation Forest stock anomaly detection
- Local RAG assistant (Mistral 7B via Ollama + pgvector)
- CI/CD on GitHub Actions, monitoring with Prometheus and Grafana, integration tests with Testcontainers