SignalForge AI

Prediction markets are often seen as “the wisdom of the crowd.” But I was curious—are they actually accurate, or just confidently wrong at times? With so much reliance on forecasts in finance, politics, and technology, I wanted to explore whether these markets truly reflect reality or if hidden biases distort them. This curiosity led me to build a system that doesn’t just consume predictions—but questions them.

What it does

SignalForge AI analyzes prediction market data to evaluate accuracy, detect bias, and identify mispriced events. It compares predicted probabilities with real-world outcomes and external signals like news trends. The system highlights where markets overreact or underreact, helping users spot opportunities and make better decisions.

How I built it

I used Zerve’s AI-native platform to drive the entire workflow. Instead of manually coding everything, I started with a question and let the AI agent handle data ingestion, cleaning, and iterative analysis.

Integrated APIs like Polymarket and Metaculus Built calibration models to compare predicted vs actual outcomes Applied statistical analysis to detect inefficiencies Deployed the project as an API and interactive app

Zerve allowed continuous iteration without switching tools, making the process faster and more intuitive.

Challenges I ran into

One major challenge was dealing with inconsistent and noisy data across different platforms. Aligning probabilities with real-world outcomes required careful normalization. Another challenge was interpreting bias correctly—distinguishing between genuine signal and short-term noise wasn’t always straightforward.

What I learned

I learned that prediction markets are not perfectly efficient—they often overreact to recent events and underweight long-term trends. I also gained experience working with AI-driven development, where guiding the system effectively is just as important as coding itself.

What’s next for SignalForge AI

Next steps include improving real-time tracking, expanding to more datasets, and enhancing the API so others can integrate forecasting insights into their own applications.

Built With

• api
• learning
• machine
• models
• numpy
• python-zerve-ai-platform-polymarket-api-metaculus-data-pandas
• rest