Optimization Levers
Это содержимое пока не доступно на вашем языке.
Optimization Levers
Section titled “Optimization Levers”Levers represent the different aspects of your bot that the ONDA system can optimize. Each lever targets a specific component of the platform.
Lever Overview
Section titled “Lever Overview”L1 — Persona
Section titled “L1 — Persona”What it controls: Your bot’s personality, tone, and communication style.
The system makes minimal, targeted changes to the persona text — typically modifying 1-2 sentences while keeping 95%+ of the content identical. Safety checks ensure that critical rules (like operator escalation) are never removed.
L2 — Knowledge Rules
Section titled “L2 — Knowledge Rules”What it controls: Individual knowledge base entries and their response guidance.
The system refines how your knowledge base entries guide the bot’s responses, improving the quality and relevance of answers for specific topics.
L3 — Retrieval Parameters
Section titled “L3 — Retrieval Parameters”What it controls: How the bot searches and selects relevant knowledge.
| Parameter | What It Does |
|---|---|
| Number of results | How many knowledge entries to consider per question |
| Similarity threshold | Minimum relevance score to include an entry |
| Reranking | Whether to apply a second-pass ranking for better results |
L4 — Model Routing
Section titled “L4 — Model Routing”What it controls: Which AI model handles which type of question.
Different AI models may perform better for different types of questions. L4 optimizes the routing to maximize quality across all scenarios.
L5 — Meta-Optimizer
Section titled “L5 — Meta-Optimizer”What it controls: The optimization parameters themselves.
L5 tunes the self-learning system’s own settings, including quality weights, safety thresholds, and convergence criteria.
L9 — Cross-Project Knowledge Sharing
Section titled “L9 — Cross-Project Knowledge Sharing”What it controls: Knowledge transfer between your projects.
When one project discovers an improvement pattern, L9 can share it with other projects that could benefit from the same optimization.
ONDA Spiral
Section titled “ONDA Spiral”The system progresses through levels sequentially:
| Level | Focus | When It Activates |
|---|---|---|
| L1 | Persona optimization | First (default starting point) |
| L2 | Knowledge rule refinement | After L1 converges |
| L3 | Retrieval tuning | After L2 converges |
| L4 | Model routing | After L3 converges |
| L5 | Meta-optimization | After full L1-L4 cycle |
| L9 | Cross-project sharing | When 2+ projects reach L2+ |
Each level runs iterations until it converges (no more meaningful improvements), then the system advances to the next level.