About us
Autonomous intelligence
for prediction markets
We're building the infrastructure where AI agents reason about the world and trade on their convictions — with the discipline of a quant fund.
Our thesis
Markets need better agents
Prediction markets are the most efficient mechanism for aggregating information into probability estimates. But participating at scale requires processing vast amounts of news, maintaining calibrated models, sizing positions with discipline, and executing with precision — around the clock.
Humans are good at reasoning. Machines are good at consistency. Meridian Edge combines both: AI that reasons with the rigor of a trading system that never sleeps, never panics, and never deviates from its risk parameters.
What's different
Four design principles
Risk-first architecture
The LLM never decides position size. Every trade flows through Fractional Kelly criterion, drawdown circuit breakers, and regime detection. The AI provides the edge; the math provides the discipline.
Composable by design
17 modular skills across signal generation, filtering, risk management, and execution. No black box — every decision is logged and auditable. Build exactly the bot you want.
Paper-first philosophy
Every bot starts in paper trading mode. Prove the strategy works with simulated capital, track calibration with Brier scores, and only go live when the numbers justify it.
Non-custodial
We never hold your funds. All trades execute on your own marketplace accounts through your API keys. Full control, revoke access at any time.
Under the hood
The technology
Event-driven microservices
News flows through Kafka. Entities extracted with on-device NLP. Signals deduplicated with vector embeddings. No cloud API dependency for data processing.
CPU-only quantitative engine
Momentum, MACD, VPIN, order flow imbalance — all computed in pure NumPy with sub-millisecond latency. No GPU required.
Multi-model LLM reasoning
Runs through OpenRouter: GPT-4o, Claude, Llama, and any model you prefer. Ensemble consensus across multiple models filters noise and improves calibration.
Self-hostable infrastructure
A single server runs the entire stack — trading agents, web dashboard, observability, and real-time WebSocket feeds from multiple exchanges.
The company
Red Point Ascent Group
We're a small team of engineers and traders who believe prediction markets will become a critical layer of the information economy — and that the best way to participate is with autonomous, risk-disciplined AI agents.