Most of the apps and notebooks here do great work on the agent and RAG side but flatten everything into text or markdown at the end. I'd like to propose a small new example under /apps that pairs Oracle AI Database 26ai with generative UI (OpenUI or similar), so the LLM picks the right component for each answer instead of just emitting prose.
Concretely, a few example queries against a seeded sample schema:
- "Revenue trend over the last 12 months" gets a line chart.
- "Top 5 accounts by region in Q1" gets a comparison table with drill-down.
- "Active users this month" gets a KPI card with delta versus the previous month.
- "Which contracts mention auto-renewal" gets source cards with snippets and citations.
- "What's driving the dip in March" gets a mixed view: chart plus supporting source cards.
The agent picks the query strategy (SQL, vector, hybrid) and the rendering in one pass, emitting typed UI primitives instead of strings.
This fills a gap I noticed when I went through the repo. There's a lot of agent and RAG coverage already (six RAG-flavored notebooks, three apps), but nothing that exercises the presentation layer or shows off 26ai's hybrid querying from a user-facing app. 26ai is a particularly good fit because it does SQL, vector, and JSON in one engine: structured queries map naturally to charts and tables, unstructured ones to source cards, and the agent can route between them without anyone wiring up two separate stacks. It's also a workshop-shaped example, which seems relevant given /workshops is currently "Coming soon."
Rough shape: 26ai for storage and search, OCI Generative AI for the model, LangGraph or the OpenAI Agents SDK for planning (both already used in the repo), and a React frontend driven by a generative UI layer. Reuses the hybrid-search pattern from notebooks/oracle_agentic_rag_hybrid_search.ipynb. Seed scripts for the sample dataset would ship with the app so it runs end-to-end with one setup step.
For a v1 I'd scope it to a working app with seed data, at least five distinct component types across the demo queries, a short README with an architecture diagram, and a local dev path against a 26ai instance. Auth, multi-tenant, RBAC on queries, and full OCI deployment notes can come later.
A few things worth deciding before I start: naming, preferred LLM provider on OCI (Cohere via OCI, Meta Llama, configurable), whether Oracle JET is a hard frontend constraint or if React is acceptable for the generative UI side, and whether this belongs in /apps or /partners. The rendering layer is also generic enough that apps/agentic_rag could adopt it later if there's appetite.
Happy to put up a draft PR if there's interest.
Most of the apps and notebooks here do great work on the agent and RAG side but flatten everything into text or markdown at the end. I'd like to propose a small new example under
/appsthat pairs Oracle AI Database 26ai with generative UI (OpenUI or similar), so the LLM picks the right component for each answer instead of just emitting prose.Concretely, a few example queries against a seeded sample schema:
The agent picks the query strategy (SQL, vector, hybrid) and the rendering in one pass, emitting typed UI primitives instead of strings.
This fills a gap I noticed when I went through the repo. There's a lot of agent and RAG coverage already (six RAG-flavored notebooks, three apps), but nothing that exercises the presentation layer or shows off 26ai's hybrid querying from a user-facing app. 26ai is a particularly good fit because it does SQL, vector, and JSON in one engine: structured queries map naturally to charts and tables, unstructured ones to source cards, and the agent can route between them without anyone wiring up two separate stacks. It's also a workshop-shaped example, which seems relevant given
/workshopsis currently "Coming soon."Rough shape: 26ai for storage and search, OCI Generative AI for the model, LangGraph or the OpenAI Agents SDK for planning (both already used in the repo), and a React frontend driven by a generative UI layer. Reuses the hybrid-search pattern from
notebooks/oracle_agentic_rag_hybrid_search.ipynb. Seed scripts for the sample dataset would ship with the app so it runs end-to-end with one setup step.For a v1 I'd scope it to a working app with seed data, at least five distinct component types across the demo queries, a short README with an architecture diagram, and a local dev path against a 26ai instance. Auth, multi-tenant, RBAC on queries, and full OCI deployment notes can come later.
A few things worth deciding before I start: naming, preferred LLM provider on OCI (Cohere via OCI, Meta Llama, configurable), whether Oracle JET is a hard frontend constraint or if React is acceptable for the generative UI side, and whether this belongs in
/appsor/partners. The rendering layer is also generic enough thatapps/agentic_ragcould adopt it later if there's appetite.Happy to put up a draft PR if there's interest.