I built the Graph‑Based‑Data‑Modeling‑Query‑System, a FastAPI‑based Order‑to‑Cash Context Graph for the SAP dataset, ingesting the data into SQLite and exposing both deterministic handlers and an optional LLM‑to‑SQL conversational interface.
The repository name is Graph‑Based‑Data‑Modeling‑Query‑System.
The project is written in Python.
Development started 2026-03 and was last updated 2026-03.
FastAPI serves on port 8000.
SQLite was chosen for its simplicity and ability to serve as a single source of truth for both the graph and chat layers. Deterministic handlers guarantee reliable answers for required benchmark questions, while the optional LLM path (using OpenRouter, Groq, or Gemini) expands coverage for broader natural‑language queries, with strict SQL safety validation.
The graph builds primary business nodes such as Customer, Sales Order, Delivery, Billing Document, Journal Entry, Payment, Product, Plant, and Storage Location, and materializes derived flow links like ORDER_TO_DELIVERY and JOURNAL_TO_PAYMENT; confidence 1.0 is assigned to direct links and heuristic fallbacks are used only when keys are missing.
The live demo is hosted at https://jesha2-o2c-context-graph.hf.space/ and the Dockerfile enables deployment as a Hugging Face Space for a no‑card public demo.