# Short form version: * Raphtory is built from scratch in **Rust. It’s fully vectorised and multi-threded**. * You can embed us into a Notebook to do traditional ML or Risk scenarios (**e.g. people connected to a SAR or Watchlist**) * **Show:** Jupyter Notebook * We **support GraphQL as a query language** and we use that for our UI integration. * **Show:** User Interface * **List the metrics throughout** * **Paste questions in chat and say:** * Hamza already discussed these with Jeffrey and Carrie. They said you will have better detailed answer ## Some metrics: * Big data: 3.5bn transactions load in 40minutes, once in our binary Apache Arrow format, it takes 27seconds to reload. * 2-hop (neighbours of neighbours) returning 10million records takes 500ms * Small data: PageRank on 30million node graph on Mac Intel takes 2.4sec, for reference NetworkX takes 300 seconds, and graph-tools takes 4sec. Neither have the temporal, multi-layer support. And Graph-tools doesn’t even support properties. * You can embed us into a Notebook to do traditional ML or Risk scenarios (e.g. people connected to a SAR or Watchlist) * Can show a Jupyter Notebook here * Then you can pull result down using GraphQL * Show the UI (this builds appreciation of how fucking good we are, they don’t even have that nice of a UI) * Go to questions ## Questions 1. Jeffrey and Carrie mentioned 2 weeks it takes to respond given a document - is there a clear bottleneck as to why? 2. Actually takes longer - datawarehouses across the business with everyone on different schemas - they are working on datamesh to connect into everything as part of cloud migration - this is the biggest bottleneck 3. How do you currently implement risk scenarios? 4. Carrie mentioned you use NetworkX, igraph, and you pull down from HPCC. How long does this take? What analysis do you do? How many people do you have working on it? * Assumption - ETL / Graph Frames into NetworkX and iGraphs - this could all be swapped to Raphtory. * Case by case scinario - all a little adhoc at the moment 5. Carrie mentioned that once you get the analysis you only store the output and push that to prod. Meaning we don’t have to be deployed on Bank premise. That’s fine, makes both of our life easier. 6. How much data do you normally work with? (Jeffrey mentioned 300m node graph - e.g. all of US population) 7. Any ideas of how much this costs you in compute/man hours? 8. Are you currently losing out on opportunities because you can’t do certain things?