Today
Loading insights…
The five platform KPIs
agreed definitions · tap a tile for the formula and SAP sourceWatchlist
Your subscriptions, re-checked against the latest snapshot
Waiting for a decision
What matters
your KPIs against target · tap a tile for its definitionWhat is likely to go wrong
dated predictions and the basis for each
What to do next
recommended actions, most urgent first — each opens the screen that does it
Full analysis
projections, components, supply options and salesSourcing within the production zone
share of purchase spend by supplier tier, per factory
Promise vs reality
SAP standard lead time vs what suppliers actually deliver (learned from receipts)
Suppliers by tier
Three-warehouse stock balancing
XGBoost demand forecast → OR-Tools minimum-cost transfers between DC01, DC21 and DC22
Factory material readiness
BOM × unrestricted stock × confirmed purchase orders (SOP SC-104)
Supplier risk
OTIF, delays, rejects, exposure and single-source risk
Governed knowledge
Contracts, SOPs, quality and product specs — pgvector + bge-small
Quality lots
Shipments & delays
SAP sandbox documents
Created by 1-click execution of accepted recommendations — never a production SAP system
Audit chain
New watch
Subscribe to a metric or a saved question; it is re-checked every time you open the command center
My watches
Flowise
LangGraph
LangGraph Business Graph
validate → workflow → policy RAG → narrative → interrupt() → review queue
Deep Agents supplier research
plan · delegate to contract / quality / logistics analysts · cited report
Customer Data Intelligence Flywheel
live counts for this customerFive metrics of moat strength
accumulating data is not the same as building an advantageSix customer data moats — and what SPARC has learned from each
Platform status
Agent backends
Workbenches
ERP / SAP Setup
Configure once, consumed everywhere: activating a connection switches this customer's SAP snapshot inside FastAPI, so the business engine, dashboards, Flowise tools and LangGraph tools all use it from the next call. Tests and syncs never activate anything; secrets stay on the server; managers make changes; every action is audited.
RAG Tuning · Company RAG Search
Tunes the real retrieval behind the SCInsight chatflow: the governed policy search (SCInsight API) and, on publish, the chatflow's own RAG DB in Flowise. Open to every role; publish and roll back are reserved for managers. Keys stay on the server.
Retrieval path (audited)
Backend apps that run the POC
Three Python backend services from the same Docker image (aurelia-scm-app:local), defined in docker-compose.yml. One FastAPI app serves both APIs (the SCInsight API is a router of it at /scinsight).
| Service | What it is | Port | Started with | Code |
|---|---|---|---|---|
| api (FastAPI) | One FastAPI app: the SPARC / Aurelia gateway at the root (dashboard, SAP OData mock, analytics, agents, approvals, /v1/embeddings for the Flowise RAG, these pages) and the SCInsight API routes at /scinsight, the governed backend of the current POC that every Flowise SCInsight tool calls (synthetic SAP data, calculations, policy RAG, recommendations awaiting approval, outcome checks, audit; own keys, roles and database) | 8000 (one Swagger: /docs) | the image's default uvicorn backend.app.main:app | backend/app/main.py includes the router of backend/scinsight/app.py |
| langgraph | LangGraph Server: runs the LangGraph graphs listed in langgraph.json, including the LangGraph copies of the SCInsight Flowise flows | 8123 | langgraph dev | backend/app/graphs.py, backend/scinsight/lg/ |
| mcp | MCP server: read-only SAP, analytics and RAG tools for MCP clients | 8002 | python -m backend.app.mcp_server | backend/app/mcp_server.py |
What they depend on, and how a chat question flows
| Service | Role | Port |
|---|---|---|
| flowise (3.1.2) | Chat bot, Agentflows, Document Store; calls the SCInsight API | 3101 |
| postgres (pgvector, pg17) | Database for the apps, the RAG vectors and the Flowise RAG DB | 55432 (localhost only) |
| Qwen LLM | Local demo: runs in Ollama on the host, outside Docker; ollama-relay forwards it to the containers (the ollama / qwen compose services are not started in this mode) | 11434 → 11435 |
| langfuse + langfuse-worker, with clickhouse, redis and objectstore (SeaweedFS) | Tracing of LLM and agent runs | 3103 |
| prometheus + grafana | Metrics and dashboards | 9090 / 3002 |
| demo-edge (Caddy) | Local proxy for the demo URLs | 3000 / 3003 (localhost only) |
Flowise chat → SC Agent Manager Agentflow → SCInsight API (/scinsight on :8000), which reads SAP data from Postgres and does the calculations, with Qwen explaining the result. The same app provides the embeddings for the Flowise RAG DB and hosts the dashboard.
Select a file to see its full code.