Verita

• 

  Logo

• 

  Logic Diagram

Inspiration

Seeing too many friends and family members lose money primarily due to overlooking their own cognitive and emotional biases or even acting under their influence all while being aware of them. Even those that spend a lot of time researching their positions and understanding the market fall into these traps. For this exact reason, retail traders are called "dumb money".

What it does

Verita detects your personal trading biases in real-time by analyzing the discrepancy between what you say you'll do and what you actually do. It builds a behavioral profile using Honcho's memory system, compares new trades against your historical patterns, and warns you when you're about to repeat costly mistakes.

How we built it

Vibe-coded using Replit. Backend logic verified through Jupyter

Challenges we ran into

Choosing + integrating an LLM with financial context to run in tandem with Honcho

Accomplishments that we're proud of

Leveraging Honcho's memory + reasoning capabilities on an application where those two things are uniquely needed + has actual real world utility. It's pretty cool that we can maintain a living behavioral profile that evolves with each trade, enabling increasingly accurate bias detection personalized to each trader's specific patterns.

Workflow

Tldr; Trader enters a trade idea → Honcho analyzes for bias → Shows bias warning where+ correction → User decides (proceed/modify/cancel) → System tracks decision → Later: Calculate actual prevented loss → Flowglad charges % of prevented loss

Process O/V 1) Users use this platform to track and evaluate their trades - they can link their trading institution 2) We use Honcho to create a profile on a user’s trading habits, and communicate between FinGPT (open source trading LLM used by many day traders), Honcho’s profile on the user, and the user’s historical + ongoing trading data. 3) Targeted queries against the user’s Honcho profile and historical trades evaluate a proposed trade for any biases. The system identifies discrepancies between stated intentions / reasoning ("will risk 2% max") and actual behavior ($5500 = 3.5% risk). Each bias detected is quantified and linked to historical outcomes from the trader's profile a) Historical Pattern Check: Compares current trade timing/size against past performance b) Language Pattern Matching: Analyzes if reasoning matches previous losing trades' language c) Size vs Success Analysis: Checks if position size deviates from profitable baseline d) Confidence Calibration: Evaluates if stated confidence aligns with actual risk taken (claims 2% risk, takes 3.5%) e) Decision Recommendation: Synthesizes all signals to provide specific guidance 4) Flowglad allows dynamic pricing based on how much the user saved from changing their trade based on recommendation.

Built With

• flowglad
• honcho