learning

Documentation

Index

• func AdjustRuntimeConfidence(baseConf float64, hasPreference bool, providerBias float64, isTimeMatch bool) float64
• func CalculatePreferenceConfidence(successCount, totalCount int, avgQuality float64) float64
• type AnalysisResult
• type LearningEngine
  • func NewLearningEngine(cfg *config.LearningConfig, mem memory.MemoryManager) (*LearningEngine, error)
  • func (le *LearningEngine) AnalyzeAll()
  • func (le *LearningEngine) AnalyzeUser(ctx context.Context, apiKeyHash string) (*AnalysisResult, error)
  • func (le *LearningEngine) ApplyPreferences(model *PreferenceModel) error
  • func (le *LearningEngine) Start()
  • func (le *LearningEngine) Stop()
• type ModelPreference
• type PreferenceModel
• type TimePattern

Functions

func AdjustRuntimeConfidence

func AdjustRuntimeConfidence(baseConf float64, hasPreference bool, providerBias float64, isTimeMatch bool) float64

AdjustRuntimeConfidence adjusts a hypothetical runtime confidence based on learned factors. This implements the design doc formula: Base + Pref Boost + Bias + Time

func CalculatePreferenceConfidence

func CalculatePreferenceConfidence(successCount, totalCount int, avgQuality float64) float64

CalculatePreferenceConfidence computes the confidence score for a learned preference. It considers success rate, sample size, and consistency.

Types

type AnalysisResult

type AnalysisResult struct {
	UserID           string
	Timestamp        time.Time
	RequestsAnalyzed int
	NewPreferences   *PreferenceModel
	Suggestions      []string
}

AnalysisResult holds the result of a history analysis run.

type LearningEngine

type LearningEngine struct {
	// contains filtered or unexported fields
}

LearningEngine orchestrates learning from routing history.

func NewLearningEngine

func NewLearningEngine(cfg *config.LearningConfig, mem memory.MemoryManager) (*LearningEngine, error)

NewLearningEngine creates a new learning engine.

func (*LearningEngine) AnalyzeAll

func (le *LearningEngine) AnalyzeAll()

AnalyzeAll triggers analysis for all known users.

func (*LearningEngine) AnalyzeUser

func (le *LearningEngine) AnalyzeUser(ctx context.Context, apiKeyHash string) (*AnalysisResult, error)

AnalyzeUser performs deep analysis on a specific user's history.

func (*LearningEngine) ApplyPreferences

func (le *LearningEngine) ApplyPreferences(model *PreferenceModel) error

ApplyPreferences applies learned preferences to the memory system.

func (*LearningEngine) Start

func (le *LearningEngine) Start()

Start starts the background analysis routine.

func (*LearningEngine) Stop

func (le *LearningEngine) Stop()

Stop stops the background analysis.

type ModelPreference

type ModelPreference struct {
	Model       string  `json:"model"`
	Confidence  float64 `json:"confidence"`
	UsageCount  int     `json:"usage_count"`
	SuccessRate float64 `json:"success_rate"`
}

ModelPreference represents a specific preference for a model given an intent.

type PreferenceModel

type PreferenceModel struct {
	UserID           string                      `json:"user_id"`
	LastUpdated      time.Time                   `json:"last_updated"`
	TotalRequests    int                         `json:"total_requests"`
	ModelPreferences map[string]*ModelPreference `json:"model_preferences"` // Key: Intent
	ProviderBias     map[string]float64          `json:"provider_bias"`     // Key: Provider, Value: -1.0 to 1.0
	TimePatterns     map[string]*TimePattern     `json:"time_patterns"`     // Key: "weekday" or "weekend"
}

PreferenceModel represents a learned model of user preferences.

type TimePattern

type TimePattern struct {
	// Simple hourly distribution (0-23)
	HourlyUsage map[int]int `json:"hourly_usage"`
	// Peak hours where specific intents are more common
	PeakIntents map[int]string `json:"peak_intents"`
}

TimePattern represents usage patterns based on time of day.