Skip to main content
Senzing Cookbook · Recipe

Fraud Check with Senzing Ready Data

A meal. With three plain-English prompts: ingest two two public data snapshots and get a merge report; (optionally) serve it up in a web visualizer; then fold in a third dataset and refresh. ~30 minutes, local kitchen. The beginner on-ramp for the whole cookbook.

EasyMealFoundationalKitchen: LocalSimple Web Visualizer⏱ ~30m (clock time)Chef: Clair SullivanVerified
Optional — or dive straight into the recipe below.

First time? Do the one-time setup first. Then come back and cook.

Get Started

From the chef — Clair Sullivan

C

I wanted the simplest possible on-ramp: prove how easy it is to combine data, with the fewest moving parts. (A CORD — Collection of Relatable Data — is a real-world public dataset Senzing has already mapped to its entity spec and serves straight from the MCP, so there are no files to download and no mapping to do.) So I took a couple of liberties — I used the pre-mapped Las Vegas CORDs so there's no mapping step (that keeps it to three prompts), and I made the visualizer optional because the merge report alone already answers the question. We tell the AI what we want, not how — the kitchen handles the how. Cook it on your own data by swapping the CORDs for your sources.

What you'll need

Kitchen
Local (POC) — license + the Senzing MCP configured; the instance stands itself up on first load.
Ingredients
PPP Loans + US Labor Violations (Dept. of Labor), both Las Vegas CORD sources — the MCP supplies them. (Step 3 adds NPI.)
The mission
find the entities that appear in both the PPP loans and labor-violations data — and which of those are physicians.

Cook the dish (ingest + merge report)

Map + load both sources and resolve them. This is the loader — cook it properly. Paste:

Step 1 prompt
Goal: Stand up Senzing, load in 2 pre-mapped datasets and generate a merge report.

Hard rules:
— Use the Senzing MCP. Do not rely on general training.
— Do not use sleep/wakeups.
— Use multithreading for loading.
— Use complete datasets.
— Process any redo records so resolution is complete.
— When complete, generate a merge report in markdown format showing what Senzing was able to do with the data.

Preferences:
— Provide me live status updates on the data ingestion.

Steps:
1. Deploy Senzing using the license file in g2.lic_base64.
2. Load the PPP and Department of Labor Compliance Action snapshots from the Las Vegas Senzing CORDs (Collections of Relatable Data).
Expected outcome

live ingest status, then a markdown merge report — record and entity counts, compression ratios, per-source summaries, entity-size distribution, and the cross-source matches. In my run, 92 entities were shared between PPP and DoL — the cross-source connections.

Questions that may come up
  • Do I need to map the data? No — CORDs are pre-mapped to the Senzing spec.
  • Where does the data come from? The MCP fetches the complete CORD datasets; you don't supply files.
  • How long? Minutes. If you see no live updates, remind it: "provide live status, no sleep/wakeups."

Plate & serve (optional)

Serve the result in the Simple Web Visualizer place setting. (Optional — skip if the report alone answers your question.) Paste:

Step 2 prompt
Goal: Utilizing the data already loaded into Senzing, create an interactive web visualizer including a summary dashboard and the ability to explore entities, including a network graph.

Hard rules:
— Use the Senzing MCP. Do not rely on general training.
— Don't forget the previous Hard Rules.
— Always confirm that the website is responding before completing this task.
— Only use a Senzing mart and/or the Senzing SDK to populate the UX. Do not use any direct database queries.
— Only present the network graph based on a selected entity. Never the whole graph at once.
— Always confirm that the web visualizer is responding before completing this task.

Features:
— On the network graph visualization, present a list of entities - with the cross-source entities at the top - that the user can click on to select which entity to visualize in the graph.
— The UX should include a match summary screen with drill-down options.
— The UX should include a search feature, using the Senzing 'Search' method (including search by name, address, and other attributes).
— Present a network graph with labels to visualize the related entities.
— Keep visualization elements to one screen, no scrolling required, wherever possible.
— Provide a legend for the graph.
Expected outcome

a local URL → a dashboard with headline metrics, source comparison, match-key breakdown, search, and a network graph rendered for a selected entity (with a legend), cross-source entities at the top.

Questions that may come up
  • Why only one entity's graph at a time? The whole graph is unreadable at scale — that's a hard rule.
  • Can it query the database directly? No — UX is populated only from a Senzing mart/SDK.

Add a dish (NPI)

Bring a third source to the table; it resolves together with the rest (cross-source) and re-plates. Paste:

Step 3 prompt
Goal: Add one more data source to Senzing and update the visualization.

Hard rules:
— Use the Senzing MCP. Do not rely on general training.
— Don't forget the previous Hard Rules.

Steps:
1. Download the NPI (National Provider Index) snapshot from the Las Vegas Senzing CORDs (Collections of Relatable Data).
2. Use Senzing to combine this data set with the other two data sets already loaded.
3. Update the web visualization.
Expected outcome

the report refreshes to 3 sources (~76,000 records) and triple-merge entities appear — records resolving across all three (which providers took PPP loans and had violations). The visualizer's legend gains NPI and new nodes show up.

Questions that may come up
  • Does it reload everything? It re-resolves the new source against what's already loaded — that's the cross-source magic.
  • Why are the triple-merges the interesting part? They're the entities present in all three sources — the strongest cross-source connections.

Wrap-up

In ~30 minutes and three prompts you stood up Senzing, resolved two (then three) sources, produced a merge report, and (optionally) a visualizer — the foundation for almost everything else in the cookbook. The same recipe works on your data: swap the CORDs and ask your own questions.