Oracle APEX and OCI Generative AI, Build an AI App

Oracle APEX is Oracle's low-code platform for building data-driven web applications on Oracle Database. OCI Generative AI is Oracle's managed service for large language models. In this video we put them together and build a small AI-assisted application from start to finish.

We create an APEX app from a table of demo supplier queries, connect the workspace to OCI Generative AI using a credential, add an AI assistant that answers questions about the data, and use AI to draft replies that a person reviews before sending. We also cover prompts, guardrails and where to keep a human in the loop.

Egisedge builds APEX + AI low-code applications for clients. Your results will depend on scope.

Reference: Oracle APEX and OCI documentation on docs.oracle.com.
Learn more: https://egisedge.com

#OracleAPEX #OCI #GenerativeAI

Transcript

[Scene: Presenter on camera, laptop open. On-screen title: "APEX + OCI Generative AI build".]

PRESENTER: In this video we build a small AI-assisted application with Oracle APEX and OCI Generative AI. The app helps a finance team handle supplier queries. It lists the queries, answers questions about them, and drafts replies. A person always reviews a reply before it goes out.

[Scene: Architecture diagram. Left: "APEX app". Middle: "Oracle Database – supplier queries table". Right: "OCI Generative AI". Arrows show the app reading data and sending prompts.]

PRESENTER: Here is the design. APEX runs on Oracle Database. Our data sits in one table. When the user asks the assistant something, APEX sends a prompt to OCI Generative AI and shows the response. We use demo data only.

[Scene: Chapter card, "Create the APEX app".]

PRESENTER: Start in the APEX workspace. We have a table called supplier queries, with columns for supplier, subject, message, received date and status.

[Scene: Screen recording. App Builder, Create Application. Add an interactive report page and a form page based on the table. Run the app.]

PRESENTER: Use Create Application and base the pages on the table. APEX gives us a searchable report and a form in a few clicks. This is where low code saves time. The standard screens are generated, so we spend our effort on the business logic.

[Scene: Show the running app. Filter the report by status "Open".]

[Scene: Chapter card, "Connect OCI Generative AI".]

PRESENTER: Next, connect the workspace to OCI Generative AI. You need an OCI tenancy with access to the Generative AI service in a supported region, and an API key for a user or a suitable policy.

[Scene: OCI console, briefly. Show where the Generative AI service lives. Blur tenancy and user identifiers.]

PRESENTER: In APEX, store the OCI credential as a web credential. Then register OCI Generative AI as a generative AI service for the workspace, and choose the model. Test the connection before you go further.

[Scene: APEX Workspace Utilities. Show the generative AI services page and a successful connection test. Blur keys.]

PRESENTER: Keep keys out of your code. Credentials belong in the workspace, managed by an administrator.

[Scene: Chapter card, "Add the AI assistant".]

PRESENTER: Now add the assistant. APEX lets you add an AI assistant to a page that uses the service we just set up. We give it a system prompt that says who it is and what it can do.

[Scene: Page Designer. Add a button "Ask AI" that opens the assistant. Show the system prompt text area.]

PRESENTER: Our system prompt says: you help the accounts payable team with supplier queries. Answer only from the data provided. If you do not know, say so. Then we pass in the open queries as context.

[Scene: Run the app. Type: "Which open queries are about late payments?" Show the assistant listing three demo queries.]

PRESENTER: The assistant answers from the data we gave it. Notice it lists the query IDs, so the user can check the source.

[Scene: Chapter card, "Draft replies with a human check".]

PRESENTER: Next, drafting replies. On the form page we add a button called Draft reply. It sends the supplier's message and a short instruction to the model, and puts the result in an editable text field.

[Scene: Form page. Click Draft reply. A draft appears in the text area. The user edits one sentence, then clicks Approve.]

PRESENTER: The draft is only a starting point. The user edits it and clicks Approve. Only then is the status updated. We also log who approved each reply and when.

[Scene: Chapter card, "Prompts and guardrails". Checklist graphic.]

PRESENTER: A few guardrails we use on every build. Keep the context small and relevant. Tell the model to answer only from the data. Never send personal data you do not need. Log prompts and responses for review. And keep a person in the loop for anything that leaves the organisation.

[Scene: Presenter on camera.]

PRESENTER: A word on cost and speed. In Egisedge's own benchmarks, APEX with AI low code has delivered up to 70 percent cost reduction and three times faster development than traditional custom development. Those are our benchmarks, not a promise. Your results depend on scope, data and integration needs.

[Scene: Chapter card, "Recap and next steps".]

PRESENTER: To recap. We created an APEX app from a table. We connected OCI Generative AI through a web credential. We added an assistant grounded in our data. And we drafted replies with a human check. For the official reference, see the APEX and OCI documentation on docs dot oracle dot com. To talk to us about an APEX build, visit egisedge dot com.

[Scene: End card with egisedge.com.]