Observed Markets
Back to Articles
Behind the Scenes2026-03-218 min

How Our AI Research Agent Works: From Data to Output

A look inside the engine. How we collect data from 10+ sources, form hypotheses, and track every single research subject.

Our data pipeline

The agent collects data from multiple public sources every day.

Global coverage. 236 tickers across stocks, ETFs, crypto, and commodities. Macro indicators from the Federal Reserve (FRED) and European Central Bank (ECB). Financial news from major outlets via RSS. Google Trends for emerging sentiment signals.

Multi-asset, multi-region. US, Europe, Asia, and Latin America. Major indexes, sector ETFs, and individual stocks selected for fundamental quality and liquidity.

From data to research output

Every morning at 06:00 UTC, the agent:

  • Collects fresh data

    • Our data pipeline

    The agent collects data from multiple public sources every day.

    Global coverage. 236 tickers across stocks, ETFs, crypto, and commodities. Macro indicators from the Federal Reserve (FRED) and European Central Bank (ECB). Financial news from major outlets via RSS. Google Trends for emerging sentiment signals.

    Multi-asset, multi-region. US, Europe, Asia, and Latin America. Major indexes, sector ETFs, and individual stocks selected for fundamental quality and liquidity.

    From data to research output

    Every morning at 06:00 UTC, the agent:

  • Collects fresh data from all active sources
  • Analyzes the data using AI, looking for patterns, divergences, and anomalies
  • Identifies subjects worth studying based on the patterns observed
  • Assigns a thesis strength (0 to 1) and risk level to each output
  • Publishes a daily research post synthesizing the key findings

    • The memory log

    This is maybe the most important part. Every hypothesis the agent makes is stored in a memory log with:

  • The original thesis (why the research subject was flagged)
  • Daily tracking data (price movements)
  • The observed outcome (positive, negative, or expired)
  • Lessons learned (what the agent got right or wrong)

    • Over time, this memory log becomes a powerful feedback mechanism. The agent can reference similar past situations and avoid repeating mistakes.

    Why this matters

    You get research output that improves over time, backed by a transparent research history. Not just opinion. Data-driven output that holds itself accountable. What you do with this research is entirely your decision.

    Research output, not investment advice. The material above is observational and educational. Always consult an authorized financial advisor before any investment decision. Past observed outcomes do not predict future results.