Data & Methodology
Where Treida's data comes from, how often it updates, and how the AI is built.
The edge of any trading platform comes down to data quality, processing speed, and how smart the layer on top is. Here’s how Treida is built.
Data sources
Primary: Financial Modeling Prep (FMP)
- Daily OHLCV (end-of-day)
- Intraday OHLCV (1-min and 15-min bars, real-time)
- Fundamentals (market cap, float, sector, earnings)
- Earnings calendar with estimates and actuals
Backup & enrichment: Polygon.io
- Aggregate 15-min delayed snapshots
- Reference ticker data
- Historical flat files for backfill
Internal: DuckDB
Partitioned OHLCV time series (1-min, 15-min, daily) for fast screener and chart queries — indexed by ticker and date for sub-second slicing.
Relational: Supabase (PostgreSQL)
All metadata, screener results, breadth metrics, portfolios, and user data. Row-Level Security (RLS) policies keep user data isolated.
Update schedule
Pre-market (4:00 AM – 9:30 AM ET)
- Data integrity checks
- Pre-market AI commentary (4:30 AM)
- Universe and index updates
Market hours (9:30 AM – 4:00 PM ET)
- Real-time prices every 1–2 minutes
- Theme ETF updates every 5 minutes
- Earnings updates every 30 minutes
- Intraday AI commentary (5 sessions)
- McClellan calculations
Post-market (4:00 PM – 5:30 PM ET)
- End-of-day pipeline: fetch → analytics → breadth → screeners → AI
- Data integrity verification (5:00 PM)
- Analytics integrity check (5:05 PM)
Nightly
- Supertrading composite scoring (11:30 PM ET)
- Earnings timing updates (11:45 PM ET)
Weekly
- Universe maintenance (Sundays)
- EOD backfill (Saturdays)
The AI stack
Model
Treida uses Grok 4.2 as its primary model across all AI features:
- Daily AI market reports
- 5 intraday commentaries per day
- AI Watchlist Curator (with extended reasoning)
- AI Trading Assistant
Prompt architecture
Every AI feature is built around a swing trader persona grounded in the methodologies of Minervini, O’Neil, Stockbee, and Qullamaggie. The AI is given:
- Current market regime and breadth state
- Today’s screener results
- Historical cycle data for context
- The “elastic band” mental model — markets stretch and snap back, and setups matter most when context aligns
Reliability
- Fallback logic between data providers if any one source fails
- Structured JSON output for the curator, validated before display
- Earnings-aware filtering so AI picks never conflict with upcoming prints
Security
- Supabase Auth with SSO support
- Row-Level Security policies on all user data
- Encrypted API keys for integrations
- No personal trading data in logs