Collect Data
Scheduled jobs gather public data from trusted sources, such as weather observations or economic datasets.
How the systems work
Architecture
Each dashboard in Jim's Prediction Portal follows the same practical pattern: collect public data, shape it into useful features, train a model, save fresh predictions, and display the results in a clean dashboard.
Prepare Features
The raw data is cleaned and reshaped into the columns a model needs, such as recent history, calendar patterns, or source freshness.
Train Models
Models are trained locally first, tested against held-out data, and saved so they can be reused by automated prediction jobs.
Publish Predictions
Cloud jobs run on a schedule, load the latest model inputs, make predictions, and save simple JSON files for the dashboard.
Show Results
The dashboard reads the latest prediction files and presents the forecast, supporting context, and model performance in one place.
Weather Intelligence
The live pipeline
The weather project uses historical weather records for training, recent weather summaries for daily predictions, and an automated AWS job that publishes a fresh seven-day high-temperature forecast for the dashboard.
Restaurant Intelligence
The next pipeline
The restaurant project is being staged around public monthly sales and labor-market data. The first goal is a simple, understandable sales forecast that can improve as more features are added.
Why this architecture matters
It turns a machine learning model into a usable product
Repeatable
The same pattern can support weather, restaurant sales, environmental data, or future prediction ideas.
Automated
Once deployed, scheduled jobs can keep predictions fresh without manually rerunning notebooks.
Inspectable
Visitors can see the forecast, the source status, and the model performance instead of only seeing a final number.
Expandable
New dashboards can be added while keeping the same navigation, explanation style, and project structure.