Kynexis Decision Intelligence Platform
Passive behavioral intelligence for investment professionals
See the decision, not just the trade.
The shift
Most tools measure the outcome. Kynexis measures the decision behind it.
Behavioral risk is the risk a firm can least see. A trader doubles down to win back a loss, sizes up on conviction that accuracy does not support, or slides into overtrading, and usually none of it is visible until it has already cost money.
Journaling tools ask traders to report on themselves, exactly when self-reporting is least honest. Trade surveillance watches the fills, not the reasoning. Kynexis was built to close that gap.
How it works
The professional trades normally, on TradingView or any other platform.
A local vision service reads the order on-device by OCR: ticker, direction, platform.
Eighteen signals weigh the decision against the trader’s own history.
A plain-language alert appears, with the evidence that triggered it.
Only a sealed envelope, pattern and severity, ever leaves the machine.
The engine · eighteen signals
Each signal is defined from the academic literature, then tested blind against 490 synthetic traders across thirty years of real market data. Two more detectors work at the desk level, looking across a team instead of a single person.
held up cleanly in blind testing
a measurable trace, reported as behavior, not proven bias
patterns across a team, never an individual
kept, but deliberately down-weighted
Privacy by construction
Privacy is the architecture, not a setting. On an individual install the sensitive data never leaves the machine. When a firm is involved, computers exchange only a sealed envelope, and it travels up the firm’s real reporting structure so a manager sees only their own direct reports.
- Screens
- Trades and positions
- Profit and loss
- Names and tickers
- Pattern type
- Severity
- Direct reports only
- No names, no tickers
How it was validated
Tested blind. Nothing shipped on intuition.
A generator built 490 traders with biases drawn from the research and ran them across thirty years of real market data. The biases were injected as academic constructs, never as the detectors’ own rules, so the test could not quietly grade itself. Signals that did not hold up were re-scoped, relabeled more honestly, or down-weighted, never kept for show.
What I built
A scoring engine, an on-device vision service, a desktop application, and web views for firms and desks.
Envelope-only messaging over a closure-table model of a firm’s reporting structure, proven in code and covered by tests.
A synthetic-trader generator and a written report that grades every signal on real market data.
Reproducible installers for three platforms, full documentation, and a diligence data room.
Built to hand over
Kynexis was finished as a complete, self-contained asset, packaged so the whole project can change hands without entanglement.
Full source, validation harness, documentation, and a clean licence audit with no strong copyleft. Built for trade-surveillance, behavioral-analytics, and RegTech buyers.
Visit getkynexis.github.io→