Dex AI Chatbot

Meet Dex, your AI documentation assistant that truly understands your documentation.

Ask Your Docs

Instead of searching for keywords, ask Dex complex questions like:

• "How do I implement authentication using OAuth?"
• "What's the difference between webhooks and polling?"
• "Show me examples of error handling"

Dex understands context and can answer follow-up questions within the conversation.

Powered by RAG

By using Retrieval-Augmented Generation (RAG), Dex is grounded in the actual documentation content:

1. Your question is converted to a 768-dimensional embedding vector
2. Hybrid search finds relevant documentation chunks
3. Retrieved context is passed to Google Gemini 2.5 Flash
4. Dex generates a response with source citations

This eliminates hallucinations - Dex only knows what's in your documentation.

Hybrid Search

Dex uses a sophisticated hybrid search algorithm combining vector similarity and keyword matching:

finalScore = (0.7 * normalizedVectorScore) + (0.3 * normalizedKeywordScore);

Vector Search

• Query is embedded using Google's text-embedding-004 model (768 dimensions)
• Cosine similarity finds semantically similar content chunks
• Works well for conceptual and natural language queries

BM25 Keyword Search

• Classic probabilistic ranking function
• Parameters: k1=1.5, b=0.75, avgDocLength=5000
• Works well for specific terms, API names, and code references

Field Weights

Field	Weight	Rationale
Title	5x	Page titles are highly relevant
Description	2x	Summaries capture key concepts
Content	1x	Body text provides detail

Features

Source Citations

Every response includes clickable source links with relevance scores. Click a source to open the original documentation page.

Thinking Steps

Expand "Dex's Thinking" to see:

• Search queries executed
• Sources evaluated and ranked
• Reasoning process
• Why certain sources were selected

Just-in-Time Scraping

If Dex needs content from a page that hasn't been scraped yet, it automatically fetches and indexes that page during the conversation.

Suggested Topics

After greeting messages or broad questions, Dex suggests relevant topics based on the discovered documentation structure.

Related Pages

At the end of responses, Dex suggests related pages you might want to explore next.

Graph Highlighting

When Dex cites sources, those nodes pulse in the visualization, helping you see where information came from.

Chat Request/Response

Request:

{
    message: string;
    appId: string;
    searchId: string;
    conversationId?: string;
    siteName?: string;
}

Response:

{
    success: boolean;
    response: string;
    conversationId: string;
    sources: SourceCitation[];
    codeBlocks: CodeBlock[];
    messageId: string;
    sourceNodeIds: string[];     // For graph highlighting
    isGreeting: boolean;
    thinkingSteps: string[];     // Dex's reasoning
    suggestedTopics?: SuggestedTopic[];
    relatedPages?: RelatedPage[];
}

Best Practices

1. Be Specific: "How do I authenticate API requests?" works better than "authentication"
2. Use Follow-ups: Dex remembers conversation context
3. Check Sources: Always verify answers against the linked documentation
4. Trigger Deep Dives: If Dex can't find content, scrape the relevant cluster first