A .NET Aspire application that processes Wookieepedia content into a temporal knowledge graph, powers AI-assisted exploration through semantic search and 50+ agent tools, and serves it all through an interactive Blazor frontend.
Conversational agent powered by OpenAI with tool-calling across six toolkits (50+ tools):
- Charts — Bar, Line, Pie, Donut, Stacked Bar, Radar, Rose, and Time Series
- Relationship Graphs — Family trees, political hierarchies, master-apprentice lineages (force + tree layout)
- Timelines — Temporal events scoped by era, date range, or entity lifetime
- Data Tables — Cross-referenced data assembled from multiple queries
- Infobox Cards — Wiki-style profiles with side-by-side comparison
- Research Articles — Formatted markdown with source citations from semantic search
All visualizations are grounded in real tool call results — no data is fabricated. Supports Bring Your Own Key (BYOK) for users who supply their own OpenAI API key. The Ask page features AI-generated dynamic example questions refreshed weekly from the knowledge graph.
AI-powered vector search over 800,000+ article passages using OpenAI embeddings and MongoDB Atlas Vector Search. Finds content by meaning, not keywords — ask "why did Anakin turn to the dark side?" and get answers from actual article text with deep-linked source references. Available in Ask AI and the Galaxy Map.
166,000+ entities and 694,000+ relationships extracted deterministically from Wookieepedia infobox data. Rich temporal facets track entity lifecycles across six semantic dimensions: lifespan, conflict, institutional, construction, creation, and publication. Two calendar systems (Galactic BBY/ABY and real-world CE).
Interactive graph visualization with force-directed and hierarchical tree layouts. Browse entities, traverse relationships by label, expand nodes on double-click, navigate with breadcrumb history, and filter edges temporally with a year slider.
Unified interactive galactic map with two modes:
- Explore — Drill from regions to systems to planets. Keyword and AI semantic search to locate places by concept (e.g., "planet where Anakin lost his legs" → Mustafar).
- Timeline — Animated playback of galactic events with territory control heatmap, era navigation, faction overlays, and event density visualization.
Browse thousands of categorized events across all Star Wars eras with category filtering, year range scoping, and BBY/ABY navigation.
Searchable, sortable, paginated tables for every infobox category: characters, ships, weapons, species, planets, governments, and more.
AI-generated biographical timelines synthesized from full wiki articles and linked pages.
StarWarsData.AppHost # .NET Aspire orchestrator
StarWarsData.ApiService # ASP.NET Core API + AI agent
StarWarsData.Admin # Admin dashboard + ETL job triggers + Hangfire
StarWarsData.Frontend # Blazor Interactive Server UI
StarWarsData.Services # Business logic, toolkits, ETL, search
StarWarsData.Models # Shared entities and DTOs
StarWarsData.ServiceDefaults # OpenTelemetry, health checks, resilience
StarWarsData.MongoDbMigrations # mongosh migration scripts (run-once container)
StarWarsData.Tests # Integration tests (xUnit + Testcontainers)
StarWarsData.AgentTests # AI agent evaluation tests (MSTest)
Nine ordered ETL phases, triggered via the Aspire dashboard or Hangfire recurring jobs:
- Download Pages — Fetch pages from Wookieepedia via the MediaWiki API, parse infobox data, store in
raw.pages - Template Views — Create MongoDB views per infobox template type (Character, Planet, etc.)
- Timeline Events + Indexes — Build categorized timeline events from the knowledge graph (dual-calendar: BBY/ABY + CE), create query indexes
- Article Chunking + Embeddings — Split pages into semantic chunks, embed with OpenAI text-embedding-3-small (1536-dim), create vector search indexes
- Knowledge Graph — Deterministic extraction builds
kg.nodes+kg.edgesfrom infobox links with temporal facets and edge weights - Character Timelines — AI-generated biographical timelines synthesized from full wiki articles
- LLM Relationship Graph — Optional LLM-based relationship extraction via OpenAI Batch API (submit/check/cleanup cycle)
- Galaxy Map — Pre-computes
galaxy.yearswith territory control, event heatmaps, and trade routes from the knowledge graph - Ask Suggestions — AI agent explores the knowledge graph and generates Ask page example questions (weekly)
Recurring Hangfire jobs: Daily incremental wiki sync (03:00 UTC), daily relationship graph builder (04:00 UTC), batch submissions every 30 min, batch status checks every 5 min, daily article chunking (05:00 UTC), weekly Ask suggestions (Sundays 03:00 UTC).
Built with the Microsoft Agents AI framework, streaming via the AGUI protocol (SSE):
- Topic Guardrail — Lightweight classifier (GPT-4o-mini) rejects off-topic queries
- Tool Registry — Six composable toolkits:
- GraphRAGToolkit (12 tools) — KG traversal + semantic search
- KGAnalyticsToolkit (16 tools) — Aggregation queries for chart data
- DataExplorerToolkit (10 tools) — Raw page and infobox queries
- ComponentToolkit (8 tools) — Frontend render descriptors (charts, graphs, tables, cards)
- StarWarsWikiSearchProvider (1 tool) — Keyword search fallback
- MongoDB MCP (3 tools) — Direct find, aggregate, count via MCP sidecar
- BYOK — Users can supply their own OpenAI API key; BYOK users and admins are exempt from rate limiting
- Semantic Search — Shared SemanticSearchService used by the agent and galaxy map
- References — All render tools support source references with Wookieepedia URLs
Keycloak provides OpenID Connect authentication. Admin role bypasses rate limiting. User identity forwarded via X-User-Id/X-User-Roles headers from the Blazor frontend.
| Layer | Technology |
|---|---|
| Orchestration | .NET Aspire 13.2 |
| Backend | ASP.NET Core, .NET 10 |
| Frontend | Blazor Interactive Server, MudBlazor, D3.js |
| Database | MongoDB (Atlas Community Edition), MongoDB.Driver 3.x |
| AI | OpenAI (GPT-5-mini, GPT-5, GPT-5.4-mini, text-embedding-3-small, Batch API), Microsoft.Extensions.AI, Microsoft.Agents.AI |
| Search | MongoDB Atlas Vector Search (semantic), MongoDB $text (keyword) |
| MCP | MongoDB MCP Server (sidecar), ModelContextProtocol client |
| Migrations | mongosh scripts in a run-once container, tracked in migrations collection |
| Background Jobs | Hangfire with MongoDB persistence |
| Auth | Keycloak (OpenID Connect) |
| Observability | OpenTelemetry (traces, metrics, logs) |
All data lives in a single unified database (default starwars-dev for local, starwars for production) with namespaced collections:
| Namespace | Collections | Purpose |
|---|---|---|
raw.* |
raw.pages, raw.job_state |
Downloaded Wookieepedia pages and ETL job state |
kg.* |
kg.nodes, kg.edges, kg.labels, kg.crawl_state, kg.batch_jobs |
Knowledge graph entities, relationships, label registry, and LLM batch tracking |
search.* |
search.chunks |
Article chunks with vector embeddings (1536-dim) |
timeline.* |
timeline.{category} |
Categorized timeline events (one collection per infobox type) |
territory.* |
territory.snapshots, territory.years |
Territory control snapshots and yearly data |
galaxy.* |
galaxy.years |
Pre-computed galaxy map temporal data |
genai.* |
genai.character_timelines, genai.character_checkpoints, genai.character_progress |
AI-generated character timelines and progress tracking |
chat.* |
chat.sessions, chat.user_settings |
User chat history and BYOK API key settings |
admin.* |
admin.job_toggles |
Feature toggles for admin jobs |
suggestions.* |
suggestions.examples |
AI-generated Ask page example questions |
Hangfire uses a separate database (starwars-dev-hangfire or starwars-hangfire). MongoDB migrations are tracked in a migrations collection.
- .NET 10 SDK
- Docker (for MongoDB MCP sidecar and Testcontainers)
- MongoDB instance (Atlas Community Edition or Atlas)
- OpenAI API key
Set the following in appsettings.json or environment variables:
{
"Settings": {
"OpenAiKey": "<your-openai-api-key>",
"OpenAiModel": "gpt-5-mini",
"DatabaseName": "starwars-dev"
}
}The default database is starwars-dev for local development. Production overrides to starwars via appsettings.json or the STARWARS_DB environment variable.
cd src/StarWarsData.AppHost
dotnet runThe Aspire dashboard will show all resources. ETL pipeline phases are available as HTTP commands in the Aspire dashboard on the Admin resource.
MongoDB migrations run automatically on startup via a run-once mongo:latest container that executes mongosh migrate.js. Migrations are tracked in the migrations collection and are idempotent.
After deploying, run "Ensure All Indexes" from the Aspire dashboard (Admin resource). This chains all index creation jobs: page indexes → chunk indexes → vector search index → KG graph indexes. The vector search index is required for semantic search to work.
This project is built and maintained as a solo effort. Significant time and cost go into running AI models during development and for the live application:
- Claude (Anthropic) — used extensively for development via Claude Code
- OpenAI (GPT-5-mini, GPT-4o-mini) — powers the in-app AI assistant, relationship extraction, and embeddings
If you find this project useful or interesting:
