feat: Ollama AI expert commentary integration

- OllamaClient utility for llama3.2:3b API calls (timeout 30s, non-fatal)
- OllamaCommentary service builds structured Turkish prompt from prediction data
- PredictionsService enriches response with ai_expert_commentary field
- Frontend prediction-card displays AI commentary section above match_commentary

src/modules/predictions/predictions.controller.ts

@@ -4,6 +4,7 @@ import {
   Post,
   Body,
   Param,
+  Query,
   HttpCode,
   HttpStatus,
   NotFoundException,
@@ -111,6 +112,7 @@ export class PredictionsController {
   async getPrediction(
     @Param("matchId") matchId: string,
     @CurrentUser() user: any,
+    @Query("nocache") nocache?: string,
   ): Promise<MatchPredictionDto> {
     const canProceed = await this.analysisService.checkUsageLimit(
       user.id,
@@ -121,13 +123,16 @@ export class PredictionsController {
       throw new ForbiddenException("ANALYSIS_LIMIT_EXCEEDED");
     }
 
-    const cached = await this.predictionsService.getCachedPrediction(matchId);
-    if (cached) {
-      // Do not record usage for cached predictions
-      return cached;
+    const bypassCache = nocache === "true" || nocache === "1";
+
+    if (!bypassCache) {
+      const cached = await this.predictionsService.getCachedPrediction(matchId);
+      if (cached) {
+        return cached;
+      }
     }
 
-    // Get from AI Engine
+    // Get from AI Engine (always fresh when nocache=true)
     const prediction = await this.predictionsService.getPredictionById(matchId);
 
     if (!prediction) {

src/modules/predictions/predictions.service.ts

@@ -31,6 +31,7 @@ import {
   AiEngineClient,
   AiEngineRequestError,
 } from "../../common/utils/ai-engine-client";
+import { generateExpertCommentary } from "../../common/utils/ollama-commentary";
 
 type ConfidenceBand = "HIGH" | "MEDIUM" | "LOW";
 
@@ -237,6 +238,19 @@ export class PredictionsService implements OnModuleInit, OnModuleDestroy {
         response.data,
         matchContext,
       );
+
+      // Generate AI expert commentary via Ollama (non-blocking, best-effort)
+      try {
+        const aiCommentary = await generateExpertCommentary(
+          response.data as unknown as Record<string, unknown>,
+        );
+        if (aiCommentary) {
+          (prediction as unknown as Record<string, unknown>).ai_expert_commentary = aiCommentary;
+        }
+      } catch {
+        // Ollama failure is non-fatal
+      }
+
       await this.recordPredictionRun(matchId, response.data);
       await this.cachePrediction(matchId, prediction);
       return prediction;