← Bac à sable
Live walkthrough

How does a Blended Finance RAG Graph actually work?

Ten real questions a blended finance practitioner faces every week — and how a typed graph would produce a source-cited answer, by traversing relations between actors, deals, instruments and evidence rather than matching text.

Synthetic illustration The retrieval mechanism, the graph schema and the query patterns below are real. The figures, deal records and source citations are invented for demonstration — no live database is connected, and nothing here should be read as data about any named institution. Each answer is labelled with the data it would require in production: some questions are answerable from public sources today, others only from an institution's own deal corpus.
Before you start

The schema at a glance

What you see. A knowledge graph of one blended finance deal: the actors who fund it, the instruments they use, the KPIs it is measured against, and the risks, standards and documents attached to it.

How it works. When you ask a question, the system does not search for words in documents: it follows the links in this graph (who funded what, with which instrument, measured how) and builds its answer only from what it finds.

Why it matters. Every answer comes with its sources, a sample size and a faithfulness check, so you can see what it is based on, and what it cannot tell you.

Full knowledge graph of a blended finance deal: actors, instruments, KPIs, context, risks, standards and evidence around the deal
?
Loading question…
 
Retrieve
Traverse
Aggregate
Generate
Evaluate
Cite
Knowledge Graph — subgraph traversal Ready — click "Ask the graph"
Deal Ukraine cyber USAID Grantor EU Grantor AFD Conc. loan DFC Guarantor responsAb. Mezzanine Paladin VC equity KSE Verifier IREX Implementer Grant 700 K€ Mezz DIB 500 K€ DFC Guar. 300 K€ Impact Eq. 600 K€ F1 Empl. 35% F2 Train. 25% F3 Econ. 25% F4 Syst. 15% Outcome M18 ✓ Ukraine FCV · UAH Cyber B2G def. Veterans wounded IRIS+ PI2764 FX Risk TCX cov. Doc Convg 2024 SFDR Art. 9
⚙

Under the hood — the graph query pattern

✓

Answer — grounded & source-cited

Faithfulness to retrieved context —

Faithfulness measures whether the answer stays inside the retrieved evidence. It does not certify that the evidence is sufficient — which is why the sample size is shown next to it, and why the system returns ranges rather than percentages below threshold.

Try another question.

The questions are real. The answers are synthetic illustrations of the retrieval mechanism. Each is tagged with the data it would need in production.