Context before prediction
The selected asset, interval, historical table, and price series establish what the model is reading before its output is introduced.
Finance · Machine learning2024Product design · Front-end · Data visualization
B3Forecast
Making a prediction useful begins with making its context visible.
A research interface for exploring Brazilian market history and turning LSTM experiments into an inspectable forecasting workflow.
01 / Challenge
What needed to become clear.
Market forecasts are easy to present as a single answer and difficult to evaluate responsibly. The interface needed to keep source data, time range, model output, and historical context in the same readable flow.
- 01A data-heavy research workflow built around Streamlit
- 02Multiple assets and selectable time ranges
- 03Dense tables and charts competing for attention
02 / Decisions
The product logic behind the interface.
One analytical reading path
Configuration stays in a compact rail while the main canvas moves from historical evidence to visual exploration and model output.
Research UI, not trading theatre
The visual system avoids simulated exchange controls and keeps the focus on traceable inputs, charts, and model experimentation.
03 / Architecture
A visible path through the system.
Asset and date-range selection
Python and Pandas time-series preparation
TensorFlow LSTM experimentation
Streamlit tables, controls, and chart views
04 / Outcome
The project became a working research surface where a visitor can inspect the data path instead of receiving an isolated prediction.