Kuber Map · the treasury of financial knowledge

A curated, learner-first knowledge map of the Financial Industry Business Ontology (FIBO) that doubles as audit-ready grounding for financial AI agents. Every concept carries its FIBO IRI as a citation; provenance (fibo vs curated) is never blurred.

▶ Open the app: https://ai-first-community.github.io/kuber-map/ — a live, installable app, not a mockup: it runs in the browser, works offline, and installs as a PWA (desktop: address-bar install; mobile: Add to Home Screen, or scan the QR on the landing page). Nothing to sign up for, nothing to pay.

▶ More demos — decision guide · guided path · compare · mobile app · grounding for agents

Which concept? — a decision guide from a plain need to the right FIBO class.

Guided path — a curated walkthrough of the loan-origination arc.

Compare — concepts side by side across FIBO's separately-governed domains.

Mobile app — the installable PWA: dark & light themes, provenance pills, FIBO citations, offline.

Built for agents — every use case exports a citable context pack; grounding is measured, not claimed.

Named for Kubera, the treasurer and god of wealth. Inspired by the Bodhi Map approach: reveal how concepts connect, not just what they mean.

---

Why this exists

Every FinTech wants AI in production; few can trust it there. A large language model is fluent but not grounded — it talks about a mortgage or a beneficial owner convincingly, but approximately, and in finance approximately-right is wrong. Kuber Map gives AI the precise, shared, auditable language of the business: the industry's own standard ontology (FIBO), made usable and packaged as grounding context an agent can cite. The full thesis is in docs/Vision.md.

FIBO is a formal, exhaustive ontology built for modelers. Kuber Map reshapes it into three things it isn't today:

1. A teachable map — a hand-picked "core" view with learner-friendly framing, where every other FIBO explorer renders the whole thing exhaustively for experts.
2. A cross-domain bridge layer — provenance-tagged links across FIBO's separately-governed domains, offered back to the EDM Council.
3. Agent grounding — per-use-case context packs that give financial AI agents accurate semantics with a FIBO provenance trail for audit.

The value, measured

Grounding is not a claim here — it is measured. The same agent answers the same benchmark with the curated FIBO context pack vs without it, scored deterministically (no LLM judge):

	Ungrounded	Grounded	across
Accuracy	47.5%	92.8% (+45.3 pt)	263 questions
Auditable (cites a real FIBO IRI)	0%	97.0%	5 use cases
Hallucinated citations	55%	0%	gpt-4o-mini

The lift is domain- and model-robust — a stronger model does not close the gap. Full write-up: SPIKE_RESULTS.md.

What's in the box

• The map — an interactive Cytoscape view (index.html landing, app.html desktop graph, m.html mobile app), all driven by one generated js/data.js. Ten FIBO domains + Commons (3,104 concepts, 6,676 relations), a curated core of 284 across 5 use cases (loan origination, KYC, securities, regulatory reporting, derivatives) with a use-case lens, and 19 validated cross-domain bridges.
• Context packs (export/) — each use case's grounding closure as pack.json (RAG), context.md (prompt injection), a self-contained OKF slice, plus a stdlib MCP retrieval server (etl/mcp_server.py). See the wiki For AI Teams.
• The bridge contribution (contrib/) — the 19 bridges packaged as an EDM Council proposal (methodology doc + RDF/Turtle), plus outreach copy and a one-page brief. Start at contrib/README.md — the single hub for everything EDM-Council-related.

A note on the edge counts. You'll see three, describing the same graph at different stages: 6,676 typed relations extracted from FIBO, 6,687 edges in the interactive map (those + the 19 curated bridges + a few guided-tour path edges), and 6,896 frontmatter edges in the OKF bundle (including cross-cluster targets awaiting their domain).

Quick start

make setup      # create venv, install deps
make fibo       # fetch the pinned FIBO source        -> fibo-source/
make commons    # fetch the Commons upper ontology     -> commons-source/
make all        # extract -> build OKF bundle -> validate
make map        # OKF bundle -> js/data.js for the map
make check      # quality gate: ruff + pytest + validate + attribution guard

The generated knowledge/ bundle and js/data.js are committed, so the map is browsable without a rebuild — just serve the folder over http (python3 -m http.server) and open index.html. Requires Python 3.11+, Node (for the map build), and git. Full walkthroughs: Getting Started · Architecture (with diagrams).

Documentation

The wiki is the home for detail: Vision & Philosophy · Use Cases · Authoring a Use Case · For AI Teams · Value Proof · Cross-Domain Bridges.

Contributing

Contributions are welcome, and this project values grounded, verifiable work over speed. Good first contributions:

• Add a use case — a curated slice for a real financial task. It's spec-driven (a JSON file under curation/usecases/), no code change; the tooling verifies every id against FIBO. Step by step: Authoring a Use Case.
• Propose a cross-domain bridge — a link FIBO doesn't draw, with a rationale + citation. The gate rejects any bridge that isn't grounded or that duplicates FIBO.
• Improve the map or the mobile app — UX, accessibility, performance (desktop and m.html are separate UIs over the same data).
• Report a FIBO fidelity issue — if the map misrepresents FIBO, cite the source IRI.

Read CONTRIBUTING.md and the Code of Conduct first. Every PR must pass make check (CI runs it automatically), keep FIBO facts grounded in source (no invented IRIs), and keep provenance (fibo vs curated) unblurred. Use the issue templates to file bugs, use-case proposals, or fidelity reports.

Repository layout

Path	What
etl/	FIBO extraction → OKF pipeline + context-pack export + bridge export + MCP server (Python)
knowledge/	Generated OKF bundle (do not hand-edit)
curation/	Hand-authored overlays: use-case specs, core sets, bridges, definitions, examples, notes
scripts/okf.js, okf.config.js, js/, *.html	The map — desktop (app.html) + mobile (m.html) UIs
export/	Generated context packs (per use case)
contrib/	The cross-domain bridges packaged as an EDM Council proposal
eval/	Grounded-vs-ungrounded eval harness + benchmarks (5 use cases)
tests/	pytest quality gate
docs/, wiki	Documentation
PLAN.md, BACKLOG.md, CLAUDE.md	Plan, execution tracker, contributor rules

License

MIT © 2026 Sanjeev Azad — see LICENSE.

This project incorporates and redistributes content from FIBO (MIT, © EDM Council & OMG) and the OMG Commons Ontology Library (MIT, © EDM Council, OMG & Thematix Partners) — both permissive and MIT-compatible — plus vendored MIT/OFL front-end libraries. Their copyright and permission notices, and a trademark disclaimer, are in THIRD_PARTY_NOTICES.md. FIBO / EDM Council / OMG are trademarks of their owners; this is an independent, community project, not affiliated with or endorsed by them.