I ve been building a side project called GPIS, a Windows-first local AI app for geopolitical and political-risk analysis.
The first idea was honestly more flashy: AI that helps predict geopolitical events.
But after testing it, I realized that was probably the worst possible framing.
The problem is that AI models can sound very confident even when the evidence is weak. In geopolitics this is dangerous, because a nice narrative can feel like analysis, even when it is just pattern-matching with too much confidence.
GPIS analyzes geopolitical events by finding recurring patterns in 100,000+ real historical records.
No API keys. No cloud. No data leaving your computer.
You describe a current event. The system retrieves similar historical situations from the GDELT corpus, runs a self-refining agentic loop (3–7 iterations), critiques its own analysis by searching for counterexamples, validates every claim against the corpus, and generates a professional PDF report with charts and confidence scores.
