feat: log LLM metadata to question_cache_meta

Made-with: Cursor

src/services/llm/llm.service.ts

@@ -1,4 +1,5 @@
 import { z } from 'zod';
+import { createHash } from 'node:crypto';
 import { env } from '../../config/env.js';
 import type { Stack, Level, QuestionType } from '../../db/schema/enums.js';
 
@@ -57,6 +58,19 @@ export interface GenerateQuestionsInput {
   types?: QuestionType[];
 }
 
+/** Metadata for persisting to question_cache_meta (used by QuestionService) */
+export interface LlmGenerationMeta {
+  llmModel: string;
+  promptHash: string;
+  generationTimeMs: number;
+  rawResponse: unknown;
+}
+
+export interface GenerateQuestionsResult {
+  questions: GeneratedQuestion[];
+  meta: LlmGenerationMeta;
+}
+
 export class LlmService {
   private readonly config: LlmConfig;
 
@@ -75,6 +89,12 @@ export class LlmService {
   }
 
   async chat(messages: ChatMessage[]): Promise<string> {
+    const { content } = await this.chatWithMeta(messages);
+    return content;
+  }
+
+  /** Returns content and model used (for logging to question_cache_meta) */
+  async chatWithMeta(messages: ChatMessage[]): Promise<{ content: string; model: string }> {
     let lastError: Error | null = null;
 
     const modelsToTry = [this.config.model];
@@ -85,7 +105,8 @@ export class LlmService {
     for (const model of modelsToTry) {
       for (let attempt = 0; attempt <= this.config.maxRetries; attempt++) {
         try {
-          return await this.executeChat(messages, model);
+          const content = await this.executeChat(messages, model);
+          return { content, model };
         } catch (err) {
           lastError = err instanceof Error ? err : new Error('LLM request failed');
           if (attempt < this.config.maxRetries) {
@@ -149,7 +170,7 @@ export class LlmService {
     }
   }
 
-  async generateQuestions(input: GenerateQuestionsInput): Promise<GeneratedQuestion[]> {
+  async generateQuestions(input: GenerateQuestionsInput): Promise<GenerateQuestionsResult> {
     const { stack, level, count, types = QUESTION_TYPES } = input;
 
     const typeList = types.join(', ');
@@ -160,10 +181,15 @@ Rules: type must be one of: ${typeList}. For single_choice/multiple_select: opti
 
     const userPrompt = `Generate ${count} questions for stack="${stack}", level="${level}". Use types: ${typeList}.`;
 
-    const raw = await this.chat([
+    const promptForHash = systemPrompt + '\n---\n' + userPrompt;
+    const promptHash = createHash('sha256').update(promptForHash).digest('hex');
+
+    const start = Date.now();
+    const { content: raw, model } = await this.chatWithMeta([
       { role: 'system', content: systemPrompt },
       { role: 'user', content: userPrompt },
     ]);
+    const generationTimeMs = Date.now() - start;
 
     const jsonStr = extractJson(raw);
     const parsed = JSON.parse(jsonStr) as unknown;
@@ -188,7 +214,15 @@ Rules: type must be one of: ${typeList}. For single_choice/multiple_select: opti
       }
     }
 
-    return questions;
+    return {
+      questions,
+      meta: {
+        llmModel: model,
+        promptHash,
+        generationTimeMs,
+        rawResponse: parsed,
+      },
+    };
   }
 }