A precise description of what happens after you deposit capital with Uvistium. No marketing, no buzzwords, no "AI-first" rhetoric.
Three separate reinforcement-learning agents analyse market data continuously across more than 40 data sources. Each agent is trained on a specific market regime: trending, range-bound, and high-volatility. A meta-selector classifies the prevailing regime every 60 seconds and routes decisions to the appropriate agent.
Model architecture. All three agents are Proximal Policy Optimisation (PPO) implementations with an adjusted reward function that maximises risk-adjusted return (Sharpe ratio over a 60-trading-day window) with explicit penalty for drawdowns exceeding 8%. Training window: 2007 to present, with rolling walk-forward validation.
Before a signal is translated into an order, it passes through four independent risk gates. If any single gate blocks, the order is not executed — regardless of signal strength or model conviction.
Approved orders are executed through FCA-authorised Tier-1 broker partners with direct market access. No routing to payment-for-order-flow venues. Each execution is logged with a microsecond timestamp: model reason, signal confidence, expected versus realised slippage, broker, and venue.
At the close of every calendar month, the attribution engine generates a personal memo. It decomposes total return to within 0.01% across contributions from model, instrument, sector, and execution day. You see exactly where return was created — and where it was surrendered.
Why this matters structurally. Most robo-advisers give a single return figure at month end. Traditional wealth managers deliver a two-page quarterly PDF. Uvistium delivers eight to twelve pages every month that you can inspect, challenge, and discuss with your adviser.
Uvistium