Inspiration
We wanted to understand how great traders think and what habits actually separate them from everyone else. Most platforms only show profits and losses, but not decision patterns. TradePersona was built to let everyday traders compare their real behavior to the habits of top trading personas.
What it does
TradePersona lets a user upload their real trading history as a CSV file and automatically:
- extracts behavioral patterns from their trades
- computes multiple bias-related signals (risk, timing, consistency, holding behavior, win/loss patterns)
- aligns the user with known trading personas (archetypes)
- visualizes the results with radar charts and alignment bars
- generates short, personalized coaching tips ## How we built it The system is built as a simple analysis pipeline:
- A React web client uploads a large CSV file.
- A Node + Express backend streams and parses the file.
- We extract behavioral features such as: average holding time, trade frequency, position size and risk concentration, win/loss and timing patterns
- These features are converted into bias scores and aligned against predefined trading personas.
- We use Gemini AI API to turn the numeric results into short, human-readable coaching explanations.
- Trade data, personas, and alignment results are stored and queried from a cloud data warehouse.
- The frontend renders radar charts, timelines, alignment bars, and coaching tips. ## Challenges we ran into Handling large CSV files: we had to make sure uploads and parsing worked reliably for real trade histories, not just small demo files. Separating core logic from AI usage: we had to clearly separate our own feature extraction and bias scoring from the AI-generated explanations, so that the core analysis stayed deterministic and explainable. Frontend integration issues: we ran into several problems with file uploads, fetch errors, and chart rendering while wiring the backend pipeline to the UI. Designing meaningful personas: defining archetypes that were distinct, realistic, and visually comparable required multiple iterations. ## Accomplishments that we're proud of Successfully parsing and analyzing large trade exports. Building a full end-to-end pipeline from raw data to behavioural insights. Creating side-by-side persona comparisons with clear visual differences. Making the system produce useful coaching explanations instead of just raw scores. ## What we learned How to design a data pipeline for real, messy user data instead of idealized datasets. How to extract meaningful behavioral signals from transactional logs. How to combine deterministic analytics with generative AI in a responsible way. How important clear visualization is when presenting complex behavioral metrics. ## What's next for TradePersona Add more behavioral features such as market regime sensitivity and entry/exit timing quality. Improve persona models using larger historical datasets. Track changes in a user’s behavior over time and show progress. Add personalized improvement goals and alerts based on detected biases.
Log in or sign up for Devpost to join the conversation.