The AI Brain
The deterministic engine that scores, plans, and adapts — and why it gives you a real answer with or without the model.
5 min read
Rules first, model second
The number behind every recommendation is computed by a deterministic rules engine, not guessed by a language model. The model only ever dresses the result in nicer prose. That means the decision you see is reproducible, explainable, and identical whether or not the AI service is reachable — Zovik never shows "AI is down" for your daily plan.
How a task is scored
Each task gets a 0–100 productivity score, a weighted blend of five factors: goal impact (35%), urgency (25%), deadline pressure (20%), your priority (15%), and effort (5%, a quick-win tie-breaker). The weights encode a point of view: work that serves an important goal and is genuinely urgent should win.
Adaptive scoring
The weights themselves learn. If your recent work keeps drifting off your goals, goal impact is weighted up. If deadlines keep slipping, the deadline clock is weighted up. Adjustments are small, capped, and only kick in once there's enough history — a new user gets the hand-tuned defaults, not a ranking reshaped by two data points.
Overload and patterns
The Brain also watches your planned load versus realistic capacity and flags days that won't fit, suggesting what to move. And it learns your patterns — the categories you tend to do on certain days, your true completion rate — feeding both the plan and your weekly review.
The rules engine is also the fallback: if the model is unavailable, you still get a full, sensible plan — just written in a plainer voice.