Concept demo · not a commercial product

About Prodicto

Prodicto is a concept project: an experiment in what forecasting could feel like if you could watch a whole population think about your question. It is a portfolio piece, not a product or a service.

How it works

1. Find the place. The question is matched against about 190 countries, 190 cities and a few regions. No place means the whole world.
2. Sample who lives there. People are drawn in proportion to where the population actually lives: big cities by metro size, the rest from towns and the countryside, ages from each country's median age. Figures are approximate 2024 values.
3. Everyone reacts. An AI model answers as each simulated person and gives their chance of YES. You watch the answers arrive on the map.
4. The verdict. The crowd's average is compared with a separate analyst-style estimate that starts from base rates.

The people are simulated by a language model. They are not real survey answers, and nobody has yet measured how accurate these forecasts are. Treat the results as an interesting experiment.

Built with

Python, FastAPI, scikit-learn, statsmodels and pandas on the server; plain JavaScript, canvas and Chart.js in the browser. Any OpenAI-compatible AI model can be connected, including open-weight models such as Qwen and Llama.

Open-source credits

• Chart.js, MIT License
• topojson-client and world-atlas country shapes, ISC License; map data from Natural Earth, public domain
• Inter and Space Grotesk typefaces, SIL Open Font License 1.1 (Inter, Space Grotesk)
• FastAPI, Uvicorn, pandas, NumPy, scikit-learn, statsmodels, HTTPX and Pillow, under their MIT and BSD-style licenses

Prodicto itself is released under the MIT License.