Spaces:
Paused
Paused
File size: 14,071 Bytes
529090e 66f3b51 529090e 66f3b51 529090e 66f3b51 529090e 66f3b51 529090e 66f3b51 529090e 66f3b51 529090e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 | /**
* Decision Engine - AI-Powered Source Selection
*
* Makes intelligent decisions about which data source to use
* based on learned patterns, current health, context, AND semantic relevance
*/
import { CognitiveMemory } from '../memory/CognitiveMemory.js';
import { getEmbeddingService } from '../../services/embeddings/EmbeddingService.js';
import { logger } from '../../utils/logger.js';
export interface QueryIntent {
type: string;
domain: string;
operation: string;
params: any;
priority?: 'low' | 'normal' | 'high';
freshness?: 'stale' | 'normal' | 'realtime';
}
export interface DataSource {
name: string;
type: string;
capabilities: string[];
description?: string; // Added description for semantic matching
isHealthy: () => Promise<boolean>;
estimatedLatency: number;
costPerQuery: number;
query?: (operation: string, params: any) => Promise<any>; // Optional query method
}
export interface SourceScore {
source: DataSource;
score: number;
breakdown: {
performance: number;
reliability: number;
cost: number;
freshness: number;
history: number;
semantic: number;
};
reasoning: string;
}
export interface DecisionResult {
selectedSource: DataSource;
score: number;
confidence: number;
reasoning: string;
alternatives: SourceScore[];
}
export class DecisionEngine {
private memory: CognitiveMemory;
private embeddings = getEmbeddingService();
private sourceEmbeddings: Map<string, number[]> = new Map();
// Scoring weights (can be tuned based on priority)
private weights = {
performance: 0.20,
reliability: 0.20,
cost: 0.15,
freshness: 0.05,
history: 0.10,
semantic: 0.30 // High weight for semantic relevance
};
constructor(memory: CognitiveMemory) {
this.memory = memory;
this.initializeEmbeddings().catch(err => {
logger.warn('Failed to initialize source embeddings:', err);
});
}
private async initializeEmbeddings() {
// Initialize embedding service (and GPU bridge if applicable)
await this.embeddings.initialize();
}
/**
* Analyze query intent to understand requirements
*/
async analyzeIntent(query: any): Promise<QueryIntent> {
// Extract intent from query structure
const intent: QueryIntent = {
type: query.type || 'unknown',
domain: query.domain || this.inferDomain(query),
operation: query.operation || this.inferOperation(query),
params: query.params || {},
priority: query.priority || 'normal',
freshness: query.freshness || 'normal'
};
return intent;
}
private inferOperation(query: any): string {
if (query?.type && typeof query.type === 'string' && query.type.includes('.')) {
const parts = query.type.split('.');
if (parts.length > 1) return parts[1];
}
return 'read';
}
/**
* Score all available sources for a query
*/
async scoreAllSources(
sources: DataSource[],
intent: QueryIntent
): Promise<SourceScore[]> {
// Ensure embeddings service is ready
await this.embeddings.initialize();
const scores = await Promise.all(
sources.map(source => this.scoreSource(source, intent))
);
// Sort by score descending
return scores.sort((a, b) => b.score - a.score);
}
/**
* Score a single source for this query
*/
async scoreSource(
source: DataSource,
intent: QueryIntent
): Promise<SourceScore> {
// Adjust weights based on priority
const weights = this.getWeights(intent);
// Calculate individual scores
const performance = await this.scorePerformance(source, intent);
const reliability = await this.scoreReliability(source, intent);
const cost = this.scoreCost(source, intent);
const freshness = this.scoreFreshness(source, intent);
const history = await this.scoreHistory(source, intent);
const semantic = await this.scoreSemanticRelevance(source, intent);
// Weighted total
const totalScore =
performance * weights.performance +
reliability * weights.reliability +
cost * weights.cost +
freshness * weights.freshness +
history * weights.history +
semantic * weights.semantic;
// Generate reasoning
const reasoning = this.generateReasoning({
performance,
reliability,
cost,
freshness,
history,
semantic
}, weights);
return {
source,
score: totalScore,
breakdown: {
performance,
reliability,
cost,
freshness,
history,
semantic
},
reasoning
};
}
/**
* Make final decision from scored sources
*/
async decide(
sources: DataSource[],
intent: QueryIntent
): Promise<DecisionResult> {
const scored = await this.scoreAllSources(sources, intent);
if (scored.length === 0) {
throw new Error('No available sources for this query');
}
const best = scored[0];
// Confidence is based on score gap between #1 and #2
const confidence = scored.length > 1
? Math.min(1.0, (best.score - scored[1].score) / 0.3 + 0.5)
: 1.0;
return {
selectedSource: best.source,
score: best.score,
confidence,
reasoning: best.reasoning,
alternatives: scored.slice(1, 4) // Top 3 alternatives
};
}
/**
* Score semantic relevance using embeddings
*/
private async scoreSemanticRelevance(
source: DataSource,
intent: QueryIntent
): Promise<number> {
try {
// 1. Get or generate embedding for source description/capabilities
let sourceVector = this.sourceEmbeddings.get(source.name);
if (!sourceVector) {
const description = `${source.name} ${source.type} ${source.capabilities.join(' ')} ${source.description || ''}`;
sourceVector = await this.embeddings.generateEmbedding(description);
this.sourceEmbeddings.set(source.name, sourceVector);
}
// 2. Generate embedding for query intent
const queryText = `${intent.type} ${intent.domain} ${intent.operation} ${JSON.stringify(intent.params)}`;
const queryVector = await this.embeddings.generateEmbedding(queryText);
// 3. Calculate cosine similarity
return this.cosineSimilarity(sourceVector, queryVector);
} catch (error) {
// Fallback to keyword matching if embedding fails
logger.warn(`Semantic scoring failed for ${source.name}:`, error);
// Simple keyword overlap fallback
const queryStr = JSON.stringify(intent).toLowerCase();
const capsStr = source.capabilities.join(' ').toLowerCase();
if (capsStr.includes(intent.type.toLowerCase())) return 0.8;
if (queryStr.includes(source.name.toLowerCase())) return 0.6;
return 0.3; // Default low relevance
}
}
private cosineSimilarity(vecA: number[], vecB: number[]): number {
if (vecA.length !== vecB.length) return 0;
let dot = 0;
let magA = 0;
let magB = 0;
for (let i = 0; i < vecA.length; i++) {
dot += vecA[i] * vecB[i];
magA += vecA[i] * vecA[i];
magB += vecB[i] * vecB[i];
}
return magA === 0 || magB === 0 ? 0 : dot / (Math.sqrt(magA) * Math.sqrt(magB));
}
/**
* Score performance (latency, throughput)
*/
private async scorePerformance(
source: DataSource,
intent: QueryIntent
): Promise<number> {
// Get average latency from memory
const avgLatency = await this.memory.getAverageLatency(source.name);
// Normalize: 0-50ms = 1.0, 500ms+ = 0.0
const latencyScore = Math.max(0, Math.min(1, 1 - (avgLatency / 500)));
// For high priority queries, penalize slow sources more
if (intent.priority === 'high' && avgLatency > 200) {
return latencyScore * 0.5;
}
return latencyScore;
}
/**
* Score reliability (uptime, success rate)
*/
private async scoreReliability(
source: DataSource,
intent: QueryIntent
): Promise<number> {
// Current health check
const isHealthy = await source.isHealthy();
if (!isHealthy) {
return 0.0; // Unhealthy source gets zero score
}
// Historical success rate
const successRate = await this.memory.getSuccessRate(
source.name,
intent.type
);
// Get failure intelligence
const intelligence = await this.memory.getSourceIntelligence(source.name);
// Penalize if there were recent failures
const recentFailurePenalty = Math.min(0.3, intelligence.recentFailures * 0.05);
return Math.max(0, successRate - recentFailurePenalty);
}
/**
* Score cost (API costs, compute)
*/
private scoreCost(source: DataSource, intent: QueryIntent): number {
const cost = source.costPerQuery || 0;
// Normalize: $0 = 1.0, $0.10+ = 0.0
const costScore = Math.max(0, Math.min(1, 1 - (cost / 0.1)));
// For low priority queries, strongly prefer free sources
if (intent.priority === 'low' && cost > 0) {
return costScore * 0.5;
}
return costScore;
}
/**
* Score data freshness
*/
private scoreFreshness(source: DataSource, intent: QueryIntent): number {
// Database sources are typically fresher than cached/file sources
const freshnessMap: Record<string, number> = {
'database': 1.0,
'api': 0.9,
'cache': 0.5,
'file': 0.3
};
const baseScore = freshnessMap[source.type] || 0.5;
// Adjust based on required freshness
if (intent.freshness === 'realtime') {
return source.type === 'database' || source.type === 'api' ? 1.0 : 0.2;
} else if (intent.freshness === 'stale') {
return 1.0; // Don't care about freshness
}
return baseScore;
}
/**
* Score based on historical patterns
*/
private async scoreHistory(
source: DataSource,
intent: QueryIntent
): Promise<number> {
// Check if this source has been successful for similar queries
const historyScore = await this.memory.getSimilarQuerySuccess(
intent.type,
intent.params
);
return historyScore;
}
/**
* Adjust weights based on query intent
*/
private getWeights(intent: QueryIntent) {
const weights = { ...this.weights };
// High priority: favor performance and reliability
if (intent.priority === 'high') {
weights.performance = 0.30;
weights.reliability = 0.30;
weights.semantic = 0.20; // Reduce semantic weight slightly
weights.cost = 0.10;
weights.freshness = 0.05;
weights.history = 0.05;
}
// Low priority: favor cost
else if (intent.priority === 'low') {
weights.performance = 0.10;
weights.reliability = 0.20;
weights.semantic = 0.20;
weights.cost = 0.40;
weights.freshness = 0.05;
weights.history = 0.05;
}
// Realtime freshness: favor databases/APIs
if (intent.freshness === 'realtime') {
weights.freshness = 0.30;
weights.performance = 0.20;
weights.semantic = 0.20;
weights.reliability = 0.20;
weights.cost = 0.10;
}
return weights;
}
/**
* Generate human-readable reasoning
*/
private generateReasoning(
breakdown: SourceScore['breakdown'],
weights: typeof this.weights
): string {
const reasons: string[] = [];
// Find strongest factor
const factors = Object.entries(breakdown).sort((a, b) => b[1] - a[1]);
const [topFactor, topScore] = factors[0];
if (topScore > 0.8) {
reasons.push(`Excellent ${topFactor} (${(topScore * 100).toFixed(0)}%)`);
} else if (topScore > 0.6) {
reasons.push(`Good ${topFactor} (${(topScore * 100).toFixed(0)}%)`);
}
// Explicitly mention semantic match if high
if (breakdown.semantic > 0.8) {
reasons.push(`Strong conceptual match`);
}
// Note any weak factors
for (const [factor, score] of factors) {
if (score < 0.3 && weights[factor as keyof typeof weights] > 0.15) {
reasons.push(`Low ${factor} (${(score * 100).toFixed(0)}%)`);
}
}
return reasons.join(', ') || 'Balanced scores across all factors';
}
/**
* Infer domain from query structure
*/
private inferDomain(query: any): string {
// Simple heuristics
if (query?.type && typeof query.type === 'string' && query.type.includes('.')) {
return query.type.split('.')[0];
}
if (query.uri?.startsWith('agents://')) return 'agents';
if (query.uri?.startsWith('security://')) return 'security';
if (query.tool?.includes('search')) return 'search';
if (query.tool?.includes('agent')) return 'agents';
return 'general';
}
}
|