Natural-language insight
Ask in plain English. The AI layer plans intent, routes to analytics and ERP API tools, and synthesizes structured answers with visuals.
AI-Native ERP
Ask your ERP anything.
Get trusted answers.
SPARC embeds a governed AI intelligence layer — LangGraph orchestration, reasoning traces, charts, and permission-scoped ERP tools. Read-only, no raw SQL.
Native AI
An intelligence layer, not a chatbot
SPARC's AI system plans, executes permission-scoped tool calls, surfaces reasoning traces, and renders metrics and charts — all from live ERP data.
Permission-scoped execution
Every tool call respects RBAC. Users only see data their role allows — same boundaries as the web UI.
Rich analytical output
Metrics, trend charts, and expandable reasoning traces. LangGraph + LangSmith — auditable, no silent writes.
How it works
From question to trusted answer
The AI layer never bypasses the transactional core. Orchestration routes through permission-scoped APIs and returns a faithful, auditable response.
User asks
Finance or ops poses a natural-language question in the web UI
AI orchestrates
LangGraph selects permission-scoped ERP API tools and builds the response
Core executes
AdonisJS validates RBAC and returns deterministic data
Answer delivered
Faithful, auditable response — read-only, no silent mutations
Trust boundaries
AI that enterprise teams can approve
Generic chatbots hallucinate SQL and leak permissions. SPARC confines AI to typed, auditable ERP tools.
No raw SQL
The AI layer uses typed ERP API tools only — never direct database access.
Read-only by design
Insights cannot silently mutate orders, stock, or invoices.
RBAC on every call
Same role permissions as the web UI — enforced server-side on each tool.
Traceable routing
LangGraph steps and LangSmith traces make every tool call auditable.
The problem
Why generic AI fails in ERP
Operational teams need fast answers from live data — but unbounded AI creates hallucinations and permission leaks in high-stakes workflows.
Today
Reporting queues
Manual exports
Ungoverned chatbots
What breaks down
-
Insight latency
Simple questions wait on reporting specialists and stale spreadsheets.
-
Unsafe AI
Hallucinated SQL and permission leaks in high-stakes workflows.
The SPARC answer
- Native AI layer — natural-language insights, reasoning traces, charts
- AI confined to trust boundaries — RBAC, auditable tool calls, no raw SQL
- Seconds from question to operational answer on live transactional data
- Full ERP core (five modules) supplies the live data the AI system queries
Target outcomes
- Seconds from question to operational answer
- Enterprise-grade assurance for AI-assisted workflows
- Cross-team alignment on stock, orders, and invoices
faster insight
data trust
safe AI
Platform
Built on a deterministic ERP core
The AI layer never replaces the core — it calls it. One monorepo with cohesive UI, transactional backend, and a separate Python process for LLM orchestration.
AI query sequence
Six bounded hops from question to insight — every call scoped by RBAC, every response auditable.
User
Web UI
AI layer
ERP API
-
Request
"What's overdue?"
-
Request
Forward query
-
Tool call
list_invoices -
Response
RBAC-filtered data
-
Synthesis
Metrics, charts & summary
-
Delivered
Insight in workspace
Data scope
Full ERP scope behind every answer
Five operational modules supply the live transactional data the AI system queries.
System Administration
Users, roles, permissions
Data Masters
Items, partners, config
Sales
Orders, invoicing, O2C
Purchasing
POs, goods receipt
Warehouse
Stock ledger, movements
Stack
AI-first architecture
AI layer
LangGraph orchestration with MCP tools and LangSmith evaluation (LLM-as-judge).
Python
FastAPI
LangGraph
MCP
LangSmith
LLM-as-judge
ERP platform
Node.js
AdonisJS v6
PostgreSQL
Lucid ORM
Inertia.js
React
TypeScript
RBAC
Methods
Agile SDLC
API-first
C4 / UML / ERD
Our Team
The graduation project team
The team behind SPARC's architecture, backend, AI, and business research.
Omar Masoud
System architect & project lead
Younes Ayman
Backend development
Khalid Hagag
Lead AI development
Omar Mohammed
AI development
Fatma Shaban
Data analysis & business research
Sarah Mohamed
Backend development
Experience
See SPARC AI in action
Watch natural-language questions flow through LangGraph orchestration to RBAC-filtered metrics, charts, and reasoning traces — grounded in live ERP data.
Demo video coming soon