Skip to main content

AI Research Engine

Meridian Themes uses an AI-powered research engine to generate automated briefings.

Overview​

The research engine:

  1. Takes theme tickers and prompts
  2. Sends queries to Perplexity via OpenRouter
  3. Receives live web research with citations
  4. Stores insights in the theme timeline

How It Works​

┌─────────────────────────────────────────────────────────┐
│                   Research Query                         │
├─────────────────────────────────────────────────────────┤
│ Theme: Copper Breakout                                   │
│ Tickers: HG=F, SCCO, FCX                                │
│ Prompt: "Analyze copper supply news from Chile"         │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│              OpenRouter → Perplexity                     │
│              (sonar-medium-online)                       │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│                  Research Output                         │
├─────────────────────────────────────────────────────────┤
│ Content: Markdown-formatted analysis                     │
│ Sources: [url1, url2, url3...]                          │
│ Generated: 2025-01-12 09:00:00                          │
└─────────────────────────────────────────────────────────┘

Perplexity Integration​

Why Perplexity?​

  • Live web search: Real-time market news
  • Citation included: Verify sources
  • Financial focus: Good at market analysis
  • Via OpenRouter: Unified API access

Model Selection​

Recommended models for research:

ModelUse Case
perplexity/sonar-medium-onlineStandard research
perplexity/sonar-large-onlineDeep analysis

Configuration​

# .env
OPENROUTER_API_KEY=sk-or-v1-xxx
RESEARCH_MODEL=perplexity/sonar-medium-online

Research Scheduling​

Frequency Options​

FrequencyUse Case
DailyActive trading themes
WeeklyLong-term positions
On-demandEvent-driven research

Scheduled Jobs​

Research can be triggered:

  • Manually via "Generate Report" button
  • Automatically via cron job
  • On-demand from theme detail page

Query Construction​

Template Structure​

function buildResearchQuery(theme: Theme): string {
  const tickerList = theme.tickers.join(', ');
  const basePrompt = theme.researchPrompts[0].prompt_text;

  return `
    Research the following financial instruments: ${tickerList}

    Focus on: ${basePrompt}

    Provide:
    1. Key recent developments
    2. Market sentiment analysis
    3. Price action summary
    4. Notable news or events

    Include source citations.
  `;
}

Prompt Engineering Tips​

Be specific about:

  • Timeframe (last 24 hours, this week)
  • Geographic focus (Chile, China, global)
  • Event types (earnings, regulatory, supply chain)
  • Output format (bullet points, summary)

Output Storage​

Theme Insights Table​

CREATE TABLE theme_insights (
  id UUID PRIMARY KEY,
  theme_id UUID REFERENCES themes(id),
  content TEXT NOT NULL,        -- Markdown report
  sources JSONB,                -- Citation URLs
  created_at TIMESTAMPTZ DEFAULT NOW()
);

Insight Object​

{
  "id": "uuid",
  "theme_id": "uuid",
  "content": "## Copper Market Update\n\n...",
  "sources": [
    "https://reuters.com/...",
    "https://bloomberg.com/..."
  ],
  "created_at": "2025-01-12T09:00:00Z"
}

Timeline Display​

Insights appear in the theme timeline:

  • Chronological order (newest first)
  • Markdown rendering
  • Expandable source list
  • Delete/regenerate options