Technical Support Use-Case

This tutorial uses a complete sample application — the curiosity-ai/technical-support repository — to walk through every layer of Curiosity Workspace on a realistic dataset: devices, parts, manufacturers, support cases, and conversation messages.

If you want a faster, framework-only tour, do Build your first enterprise AI app first; come back here when you want to see what a production-shaped app looks like end to end.

Prerequisites

• Complete the developer prerequisites.
• Clone the sample repo: git clone https://github.com/curiosity-ai/technical-support.
• A running local workspace from the Quickstart.
• An LLM/embedding provider configured under Settings → AI Settings.

Estimated time: 60–90 minutes the first time through.

What you'll learn (by example)

By the end, you will have:

• a workspace with a well-defined graph schema
• a connector that maps JSON records into nodes/edges
• search configured (text + optional embeddings)
• one or more custom endpoints implementing domain logic (similarity, search wrappers, chat orchestration)
• a custom interface tailored to your users

Step 1: Define the domain model

• Identify 3–7 core entity types (nodes).
• Identify the relationships users navigate (edges).
• Choose stable keys for each node type.

See Data Integration → Schema Design.

Step 2: Implement ingestion

• Choose ingestion approach (connector vs configured integration).
• Build the mapping into nodes/edges.
• Ensure idempotency (reruns do not duplicate).

See Data Integration → Connectors and Data Integration → Ingestion Pipelines.

Step 3: Validate graph correctness

Validate:

• node counts by type
• edge completeness
• correct keys (no duplicates)

See Reference → Graph Query Language.

Step 4: Configure text search and facets

• Choose searchable fields (titles, summaries, identifiers).
• Configure facets that match user refinement behavior.
• Tune ranking via boosts and scoping.

See Search → Text Search and Search → Ranking Tuning.

Step 5: Add embeddings and hybrid retrieval (optional)

• Enable embeddings on long, descriptive fields.
• Enable chunking for long text.
• Use hybrid search for balanced precision/recall.

See Search → Vector Search and Search → Hybrid Search.

Step 6: Add endpoints for domain logic

Create endpoints for:

• aggregates and analytics
• similarity and recommendations
• AI orchestration (retrieve → generate → store)

See APIs & Extensibility → Custom Endpoints.

Step 7: Build a user experience (optional)

Start with the default UI, then add a custom interface if you need domain-specific workflows:

• entity-centric pages (hub entities)
• curated search experiences (type-scoped + facets)
• workflow views (triage, investigation, review)

Next steps

• Operationalize the environment: Administration → Deployment and Monitoring
• Apply design patterns: Best Practices → Graph Design Patterns