About PolyGuru
Calibration over hype.
PolyGuru exists because most AI betting products measure the wrong thing. We built the tool we wished existed: one that publishes its calibration, shows its methodology, and admits when there's no edge.
Our mission
Bring serious, measurable, calibrated AI analysis to Polymarket — and make the accuracy data public so you don't have to take anyone's word for it.
Prediction markets are one of the most interesting places on the internet to apply AI. Real money, clear resolution, measurable outcomes. But they're also where the “85% accurate” marketing language kills hardest — because accuracy without calibration can hide catastrophic bet-sizing errors. We publish our numbers daily. If the reliability plot drifts, you see it before we do.
Transparency
Every resolved prediction, sortable. Reliability plot per category. We bias toward showing our misses, not hiding them.
Calibration
A 70% prediction should resolve YES 70% of the time. That's the number we optimize. Accuracy is a marketing word.
Restraint
NEUTRAL is a valid output. When evidence is thin, we match the market price and say 'no edge'. No forced signals.
Who's behind it
PolyGuru is a product of Guru Softwares Ltd, a UK-registered software shop building independent products at the intersection of AI and finance. We don't take venture funding; PolyGuru is a self-funded project that lives or dies on product quality and honest accuracy.
We believe publishing calibration data is the minimum honest bar for anyone claiming AI-powered betting intelligence. Most of our competitors refuse to do it. We'd rather lose skeptical customers who can read our calibration plot than lock them in with cherry-picked wins.
What we believe about AI + betting
Ensembles beat individuals
We run multiple AI models on every analysis and compute agreement. When they disagree, that's signal the market is hard to price — and confidence should be lower.
Ground truth beats opinion
For crypto threshold markets we pull live CoinGecko price data and inject it as TIER 1 evidence. The AI can't hallucinate a $1,800 low if it can see the real monthly low was $2,003.
Categories need different playbooks
Prediction markets are efficient → edge-hunting. Sports markets lag sharp books → confidence-matching. Applying one strategy to both loses edge.
Self-learning is mandatory
Every Sunday we retrain per-category confidence thresholds based on the week's resolved outcomes. Categories where the AI has been over-confident get stricter gates. The system corrects itself.
Say hi
Questions, feedback, press, partnerships — we read every message.