Explorer

The Explorer is Graph's hybrid AI answering engine, available at /graph/explorer. It provides a Perplexity-style conversational interface for querying the entire ResearchArk knowledge graph. Rather than browsing manually, you type a question in natural language and receive a streaming AI-generated answer that synthesizes results from multiple data sources -- Neo4j graph traversals, Weaviate vector similarity, and PostgreSQL keyword and SQL queries -- all searched in parallel.

Two Display Modes

The Explorer operates in two modes that transition automatically:

Hero mode is the landing state. A centered search bar with personalized query suggestions is displayed on a clean page. Stats badges below the search bar show the current data scale (134K+ projects, 258K+ organizations, 515K+ publications, 9,400+ opportunities). Once you submit a query, the interface transitions to compact mode.

Compact mode activates after your first query. The search bar moves to the top of the page, and the layout splits into two panels. The left panel shows the conversation thread (your queries, source indicators, and AI-generated answers). The right panel -- the Generative UI Panel -- renders inline visualizations that the AI produces as part of its response.

AI Toggle

A toggle in the search bar switches between two modes:

• AI on -- Your query is sent to the LLM service, which searches all data sources, synthesizes the results, and streams back a natural-language answer. The answer may include inline visualization directives that render as charts, tables, or statistics in the right panel.
• AI off -- Your query runs the same hybrid search across all data sources but returns raw search results without LLM synthesis. Results are displayed in a structured panel, useful when you want to see exactly what the data sources returned without AI interpretation.

Model Selection

The Explorer supports multiple Gemini models, selectable via a dropdown in the search bar. Available models depend on your account tier:

Free-tier users have access to Gemini 2.5 Flash Lite. Higher-tier models offer improved reasoning and longer context windows but consume more resources. Locked models appear with a lock indicator in the dropdown.

Hybrid Search Pipeline

When you submit a query, the Explorer searches four data sources simultaneously:

1. Keyword search (PostgreSQL) -- Full-text search across project titles, abstracts, organization names, and other text fields.
2. Vector search (Weaviate) -- Semantic similarity search using 768-dimensional EmbeddingGemma-300M embeddings. Finds conceptually related content even when the wording differs from your query.
3. Graph search (Neo4j) -- Cypher queries that traverse relationships in the knowledge graph, following links between projects, organizations, funding schemes, topics, and other entities.
4. SQL aggregation (PostgreSQL) -- Structured queries for statistical data such as funding totals, project counts, and distribution breakdowns.

Source indicators appear below your query as each data source responds, showing which systems contributed to the answer. This transparency helps you understand the provenance of the information.

Streaming Responses

Answers are streamed token by token as the model generates them. During streaming, you see:

• A pulsing indicator while data sources are being searched
• Source indicators showing which databases responded (keyword, vector, graph, SQL)
• The answer text appearing progressively in the left panel
• Visualizations rendering in the right panel as the AI emits visualization directives

Inline Visualizations

The AI can embed visualization directives within its response. These render as interactive components in the Generative UI Panel on the right side of the screen. Supported visualization types include:

• Bar charts -- Comparative data such as funding by country or projects by programme
• Pie charts -- Distribution breakdowns such as fund allocation or sector composition
• Tables -- Structured data with sortable columns
• Statistics cards -- Key metric highlights with labels and values
• Timelines -- Chronological progressions such as framework programme evolution
• Network graphs -- Relationship diagrams between entities

Visualizations are generated dynamically based on the query context. A question about funding distribution might produce a pie chart, while a question about programme history might produce a timeline.

Conversations and Follow-ups

The Explorer maintains a conversation thread within a session. After the initial answer, a follow-up input field appears. Follow-up questions carry the full conversation history to the model, enabling context-aware responses -- for example, asking "What about in Germany specifically?" after an initial question about EU-wide funding patterns.

You can clear the conversation at any time using the clear button to start fresh.

Personalized Suggestions

When signed in, the Explorer fetches personalized query suggestions based on your profile and activity. These appear as clickable chips below the search bar in hero mode, providing quick-start queries tailored to your research interests.

Related Pages

• Projects -- Browse the underlying project data that Explorer searches
• Organizations -- Browse organization records directly
• Schemes -- Explore funding scheme analytics
• Researchers -- Topic exploration and sector insights dashboards