I built GPIS, a Windows-first local AI workstation for geopolitical/political-risk analysis.
The goal is not "AI predicts the future." I actually had to move away from that framing because early local models can become overconfident or latch onto irrelevant threat categories. The current design is more conservative:
- local Ollama models;
- GDELT ingestion;
- ChromaDB retrieval;
- iterative analyst/reviewer loops;
- counterexample search;
- confidence and validity scoring;
- prompt quality gates;
- blind temporal backtesting;
- PDFreports;
It runs locally after setup and does not require cloud API keys.
I am looking for feedback on positioning and evaluation:
- Would you trust this more as an analyst workstation than as a "prediction" product?
- What would you require before using it in a company?
- Is Windows-first a mistake, or useful for nontechnical users?
Disclosure: I built it and I am preparing a paid Gumroad release. Happy to share details about the architecture and the mistakes I ran into.
Top comments (1)
Here the link if you want to try: foresightlab.gumroad.com/l/gpis-pr...