🔴 Develop an LLM-Based Application with Graph in SAP Integration Suite

Learn how the concise metadata of the new and unified Graph API makes it even quicker and easier can be utilized to develop Large Language Model (LLM) based applications. With Graph, developers access SAP-managed business data as a single semantically connected data graph using modern open standards like OData v4 and GraphQL. Get ready to revolutionize your SAP development journey!
Command Line LLM Application for Graph in Integration Suite (GitHub repo): github.com/SAP....
After the session, complete the following tutorial so you can earn points for the amazing Devtoberfest contest: developers.sap....
🛝 You can find the slides for this session and all the Devtoberfest 2023 🔴 integration track sessions here - dam.sap.com/ma...
👉Check out other sessions part of the Devtoberfest 2023 🔴 integration track - groups.communi....

[@khanna9paresh]

Nicely done and very well explained! 👍

[@jitendrasonar2787]

Excellent! Well explained…. Great efforts… keep it up the good work. Stay safe and blessed! Kudos!

[@gouravmisra2317]

Namaste sir good afternoon 🎉🎉 💜❤️ Heartiest salute sir ❤❤❤❤❤❤❤❤ Jay hind ❤️💜 Jay bharat 🙌🙌 saath ek lakshya bhed 🎉🎉

[@AlvaroBlancoRobayna]

Mmmm... are we feeding OpenAI with our business information here or by using it "just" as a translator of what the Graph API responds we can be confident that this data won't be used in any way by the LLM provider?

[@somnathpaul2020]

Yes this use case sounds risky with public API may be acceptable with Business API version of Open AI.

[@shardulsonar7784]

So in this approach, the communication to LLM happens in 2 places 1. When we send the OData metadata of the API (Which does not contain any personal data about the customers), and expect an OData query as a response, which makes the LLM act as a "Translator". 2. When we send the JSON response from the OData query to the LLM and expect a human-readable response. In the first part, we do not send any personal data to LLM, and in the second one we do send the personal data (which is fine as it is a Public Sandbox in this case), but I think we can encode the important customer information before sending it to the LLM and then decrypt while printing the human-readable response. Also, the main focus was how easy it is to get an OData query just by reading the OData metadata and not the converting of JSON to human readable string. So, in this case, even, printing the JSON is fine in this case, as it also answers your question.

[@NabheetMadan]

good stuff... To understand it LLM understadns OData metdata and when person ask something you take help of LLM to create a OData query for our Graph API's. Base on the URL which is returned via OpenAI LLM next call to Graphql is made correct?

[@shardulsonar7784]

If by "next call to Graphql is made" you mean "next call to Graph is made", then yes you are correct 🙂

[@AbhijeetKankani]

So while sending data to openai or LLM we can just send metadata and by this way we can maintain privacy of customer right?

[@shardulsonar7784]

Exactly you just need the OData metadata and not any customer/business information. you can check the example query which is sent to the LLM here: github.com/SAP-samples/graph-example-apps/blob/main/devtoberfest23/llm-graph/resources/query%20example.txt

[@khanna9paresh]

It is not advised to share confidential/sensitive information over APIs to LLMs