feat(interview): 최근 N개 세션 질문 전달해 중복 질문 회피

i3months · claude · i3months · commit d936c724aef7 · 2026-06-04T14:37:43.000+09:00
여러 번 면접 시 같은 질문이 반복되는 문제를 줄이기 위해, 질문 풀 생성
요청에 같은 유저의 최근 세션 질문을 동봉하고 AI 프롬프트가 의미 중복을
피하도록 한다. 범위는 최근 N개 세션(기본 3, interview.question-dedup.*).

- Core: InterviewSessionRepository/SessionQuestionPoolRepository 조회 추가,
  SessionQuestionsRequester 가 recentQuestions 를 payload 에 동봉.
- AI: GenerateQuestionsRequest.recent_questions + 프롬프트 회피 지시.

Co-Authored-By: Claude Opus 4.8 &lt;noreply@anthropic.com&gt;
diff --git a/ai/src/ai_server/chain/prompts/question_generation.py b/ai/src/ai_server/chain/prompts/question_generation.py
@@ -36,6 +36,11 @@
     "CS_FUNDAMENTAL·BEHAVIORAL 은 선택(없으면 빈 문자열).\n"
     "   - expected_signal: 좋은 답변이 드러내야 할 핵심(평가 관점). 한 문장.\n"
     "5. 지원자 컨텍스트의 기술/프로젝트를 구체적으로 언급하는 PROJECT_DEEP_DIVE / "
-    "TECH_CHOICE 질문을 우선합니다.\n\n"
+    "TECH_CHOICE 질문을 우선합니다.\n"
+    "6. 아래는 같은 지원자가 최근 면접에서 이미 받은 질문들입니다. 이와 "
+    "**의미상 중복되는 질문은 만들지 말고**, 새로운 주제나 다른 각도로 출제하세요:\n"
+    "---\n"
+    "{recent_questions}\n"
+    "---\n\n"
     "{format_instructions}"
 )
diff --git a/ai/src/ai_server/chain/question_generation_chain.py b/ai/src/ai_server/chain/question_generation_chain.py
@@ -18,6 +18,12 @@ class GeneratedQuestionPool(BaseModel):
     questions: list[GeneratedQuestion] = Field(default_factory=list)
 
 
+def _format_recent_questions(recent_questions: list[str] | None) -> str:
+    if not recent_questions:
+        return "(없음)"
+    return "\n".join(f"- {q}" for q in recent_questions)
+
+
 class QuestionGenerator(Protocol):
     async def generate(
         self,
@@ -26,6 +32,7 @@ async def generate(
         mode: str,
         max_questions: int,
         context: str,
+        recent_questions: list[str] | None = None,
     ) -> GeneratedQuestionPool: ...
 
 
@@ -40,13 +47,15 @@ async def generate(
         mode: str,
         max_questions: int,
         context: str,
+        recent_questions: list[str] | None = None,
     ) -> GeneratedQuestionPool:
         result = await self._chain.ainvoke(
             {
                 "job_category": job_category,
                 "mode": mode,
                 "max_questions": max_questions,
                 "context": context,
+                "recent_questions": _format_recent_questions(recent_questions),
             }
         )
         if not isinstance(result, GeneratedQuestionPool):
diff --git a/ai/src/ai_server/messaging/consumers/questions_consumer.py b/ai/src/ai_server/messaging/consumers/questions_consumer.py
@@ -90,6 +90,7 @@ async def handle(self, message: AbstractIncomingMessage) -> None:
                 mode=req.mode,
                 max_questions=effective_pool_size,
                 context=context_text,
+                recent_questions=req.recent_questions,
             )
 
             payload = QuestionPoolCallbackPayload(
diff --git a/ai/src/ai_server/model/messages/questions.py b/ai/src/ai_server/model/messages/questions.py
@@ -36,6 +36,8 @@ class GenerateQuestionsRequest(BaseModel):
     documents: list[DocumentContext] = []
     initial_question_count: int = 1
     max_questions: int = 10
+    # 같은 유저가 최근 면접에서 받은 질문들. 의미 중복 회피에 사용.
+    recent_questions: list[str] = []
 
 
 class GeneratedQuestion(BaseModel):
diff --git a/ai/tests/test_question_generation_chain.py b/ai/tests/test_question_generation_chain.py
@@ -0,0 +1,38 @@
+from __future__ import annotations
+
+from unittest.mock import AsyncMock
+
+import pytest
+
+from ai_server.chain.question_generation_chain import (
+    GeneratedQuestionPool,
+    LlmQuestionGenerator,
+    _format_recent_questions,
+)
+
+
+def test_format_recent_questions_empty():
+    assert _format_recent_questions(None) == "(없음)"
+    assert _format_recent_questions([]) == "(없음)"
+
+
+def test_format_recent_questions_bullets():
+    assert _format_recent_questions(["질문 A", "질문 B"]) == "- 질문 A\n- 질문 B"
+
+
+@pytest.mark.asyncio
+async def test_generate_forwards_formatted_recent_questions_to_chain():
+    chain = AsyncMock()
+    chain.ainvoke = AsyncMock(return_value=GeneratedQuestionPool(questions=[]))
+    generator = LlmQuestionGenerator(chain)
+
+    await generator.generate(
+        job_category="BACKEND",
+        mode="TECHNICAL",
+        max_questions=3,
+        context="ctx",
+        recent_questions=["이전 질문"],
+    )
+
+    chain_input = chain.ainvoke.call_args.args[0]
+    assert chain_input["recent_questions"] == "- 이전 질문"
diff --git a/backend/src/main/java/com/stackup/stackup/session/application/SessionQuestionsRequester.java b/backend/src/main/java/com/stackup/stackup/session/application/SessionQuestionsRequester.java
@@ -14,6 +14,8 @@
 import com.stackup.stackup.session.application.dto.GenerateQuestionsPayload;
 import com.stackup.stackup.session.application.dto.GenerateQuestionsPayload.DocumentContext;
 import com.stackup.stackup.session.application.event.SessionCreatedEvent;
+import com.stackup.stackup.session.domain.InterviewSessionRepository;
+import com.stackup.stackup.session.domain.SessionQuestionPoolRepository;
 import java.io.IOException;
 import java.io.InputStream;
 import java.nio.charset.StandardCharsets;
@@ -22,6 +24,8 @@
 import lombok.RequiredArgsConstructor;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.data.domain.PageRequest;
 import org.springframework.stereotype.Component;
 import org.springframework.transaction.annotation.Propagation;
 import org.springframework.transaction.annotation.Transactional;
@@ -44,28 +48,53 @@ public class SessionQuestionsRequester {
     private final RabbitMqProperties properties;
     private final AnalyzedDocumentRepository documentRepository;
     private final ObjectStorageClient storage;
+    private final InterviewSessionRepository sessionRepository;
+    private final SessionQuestionPoolRepository questionPoolRepository;
+
+    // 최근 몇 개 세션까지 거슬러 중복 질문을 회피할지. 0 이면 비활성.
+    @Value("${interview.question-dedup.recent-sessions:3}")
+    private int recentSessionCount;
+    // AI 에 넘길 과거 질문 최대 개수 (envelope 비대화 방지).
+    @Value("${interview.question-dedup.max-questions:30}")
+    private int maxRecentQuestions;
 
     @Transactional(propagation = Propagation.NOT_SUPPORTED)
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
     public void onSessionCreated(SessionCreatedEvent event) {
         List<DocumentContext> documents = buildDocumentContexts(event.contextDocumentIds());
         int generalCount = event.generalQuestionCount() != null
             ? event.generalQuestionCount() : DEFAULT_GENERAL_QUESTION_COUNT;
+        List<String> recentQuestions = recentQuestions(event.userId(), event.sessionId());
         GenerateQuestionsPayload payload = new GenerateQuestionsPayload(
             event.sessionId(),
             event.mode(),
             event.jobCategory(),
             documents,
             generalCount,
-            event.maxQuestions()
+            event.maxQuestions(),
+            recentQuestions
         );
         publisher.publishToAi(
             properties.routingKeys().generateQuestions(),
             payload,
             new MessageContext(event.userId(), event.sessionId(), null, null)
         );
-        log.info("generate.questions published. sessionId={}, doc_count={}, general_count={}, max={}",
-            event.sessionId(), documents.size(), generalCount, event.maxQuestions());
+        log.info("generate.questions published. sessionId={}, doc_count={}, general_count={}, max={}, recent_q={}",
+            event.sessionId(), documents.size(), generalCount, event.maxQuestions(), recentQuestions.size());
+    }
+
+    // 같은 유저의 최근 N개 세션에서 출제된 질문을 모아 AI 의 중복 회피용으로 전달.
+    private List<String> recentQuestions(Long userId, Long currentSessionId) {
+        if (recentSessionCount <= 0 || maxRecentQuestions <= 0) {
+            return List.of();
+        }
+        List<Long> sessionIds = sessionRepository.findRecentSessionIds(
+            userId, currentSessionId, PageRequest.of(0, recentSessionCount));
+        if (sessionIds.isEmpty()) {
+            return List.of();
+        }
+        return questionPoolRepository.findRecentQuestions(
+            sessionIds, PageRequest.of(0, maxRecentQuestions));
     }
 
     private List<DocumentContext> buildDocumentContexts(List<Long> documentIds) {
diff --git a/backend/src/main/java/com/stackup/stackup/session/application/dto/GenerateQuestionsPayload.java b/backend/src/main/java/com/stackup/stackup/session/application/dto/GenerateQuestionsPayload.java
@@ -12,7 +12,9 @@ public record GenerateQuestionsPayload(
     // Number of questions AI should generate for this initial callback.
     Integer initialQuestionCount,
     // Overall session question limit; not the generation batch size.
-    Integer maxQuestions
+    Integer maxQuestions,
+    // 같은 유저가 최근 면접에서 받은 질문들. AI 가 의미 중복 회피에 사용.
+    List<String> recentQuestions
 ) {
     public record DocumentContext(
         Long documentId,
diff --git a/backend/src/main/java/com/stackup/stackup/session/domain/InterviewSessionRepository.java b/backend/src/main/java/com/stackup/stackup/session/domain/InterviewSessionRepository.java
@@ -5,9 +5,19 @@
 import org.springframework.data.domain.Page;
 import org.springframework.data.domain.Pageable;
 import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.data.jpa.repository.Query;
+import org.springframework.data.repository.query.Param;
 
 public interface InterviewSessionRepository extends JpaRepository<InterviewSession, Long> {
 
+    // 중복 질문 회피용: 유저의 최근 세션 id (현재 세션 제외, 최신순).
+    @Query("select s.id from InterviewSession s "
+        + "where s.user.id = :userId and s.id <> :excludeId and s.deleted = false "
+        + "order by s.createdAt desc")
+    List<Long> findRecentSessionIds(@Param("userId") Long userId,
+                                    @Param("excludeId") Long excludeId,
+                                    Pageable pageable);
+
     List<InterviewSession> findByUser_Id(Long userId);
 
     Optional<InterviewSession> findByIdAndUser_Id(Long id, Long userId);
diff --git a/backend/src/main/java/com/stackup/stackup/session/domain/SessionQuestionPoolRepository.java b/backend/src/main/java/com/stackup/stackup/session/domain/SessionQuestionPoolRepository.java
@@ -1,12 +1,21 @@
 package com.stackup.stackup.session.domain;
 
+import java.util.List;
 import java.util.Optional;
+import org.springframework.data.domain.Pageable;
 import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.data.jpa.repository.Query;
+import org.springframework.data.repository.query.Param;
 
 public interface SessionQuestionPoolRepository extends JpaRepository<SessionQuestionPool, Long> {
 
     // 다음에 쓸 일반질문(미사용 중 idx 최솟값).
     Optional<SessionQuestionPool> findFirstBySessionIdAndUsedFalseOrderByIdxAsc(Long sessionId);
 
     long countBySessionId(Long sessionId);
+
+    // 중복 질문 회피용: 주어진 세션들에서 출제된 질문 텍스트(최신순).
+    @Query("select p.question from SessionQuestionPool p "
+        + "where p.sessionId in :sessionIds order by p.createdAt desc")
+    List<String> findRecentQuestions(@Param("sessionIds") List<Long> sessionIds, Pageable pageable);
 }
diff --git a/backend/src/main/resources/application.yml b/backend/src/main/resources/application.yml
@@ -132,3 +132,9 @@ app:
     stream-token-ttl: 60s
   cors:
     allowed-origins: ${CORS_ALLOWED_ORIGINS:https://your-domain}
+
+interview:
+  # 질문 풀 생성 시 최근 N개 세션의 질문을 AI 에 전달해 중복 회피. 0 이면 비활성.
+  question-dedup:
+    recent-sessions: ${INTERVIEW_DEDUP_RECENT_SESSIONS:3}
+    max-questions: ${INTERVIEW_DEDUP_MAX_QUESTIONS:30}
diff --git a/backend/src/test/java/com/stackup/stackup/session/application/SessionQuestionsRequesterTest.java b/backend/src/test/java/com/stackup/stackup/session/application/SessionQuestionsRequesterTest.java
@@ -13,15 +13,19 @@
 import com.stackup.stackup.document.domain.AnalyzedDocumentRepository;
 import com.stackup.stackup.session.application.dto.GenerateQuestionsPayload;
 import com.stackup.stackup.session.application.event.SessionCreatedEvent;
+import com.stackup.stackup.session.domain.InterviewSessionRepository;
 import com.stackup.stackup.session.domain.JobCategory;
 import com.stackup.stackup.session.domain.SessionMode;
+import com.stackup.stackup.session.domain.SessionQuestionPoolRepository;
 import java.util.List;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.extension.ExtendWith;
 import org.mockito.ArgumentCaptor;
 import org.mockito.InjectMocks;
 import org.mockito.Mock;
 import org.mockito.junit.jupiter.MockitoExtension;
+import org.springframework.data.domain.Pageable;
+import org.springframework.test.util.ReflectionTestUtils;
 
 @ExtendWith(MockitoExtension.class)
 class SessionQuestionsRequesterTest {
@@ -30,6 +34,8 @@ class SessionQuestionsRequesterTest {
     @Mock RabbitMqProperties properties;
     @Mock AnalyzedDocumentRepository documentRepository;
     @Mock ObjectStorageClient storage;
+    @Mock InterviewSessionRepository sessionRepository;
+    @Mock SessionQuestionPoolRepository questionPoolRepository;
     @InjectMocks SessionQuestionsRequester requester;
 
     @Test
@@ -58,6 +64,29 @@ void onSessionCreated_requestsOneInitialQuestionAndKeepsSessionLimit() {
         // generalQuestionCount(n) 만큼 생성 요청 (이벤트의 n=3).
         assertThat(payload.initialQuestionCount()).isEqualTo(3);
         assertThat(payload.maxQuestions()).isEqualTo(5);
+        // dedup 비활성(기본 0) 이면 과거 질문 없이 빈 목록.
+        assertThat(payload.recentQuestions()).isEmpty();
+    }
+
+    @Test
+    void onSessionCreated_includesRecentQuestionsForDedup() {
+        when(properties.routingKeys()).thenReturn(mockRoutingKeys());
+        ReflectionTestUtils.setField(requester, "recentSessionCount", 3);
+        ReflectionTestUtils.setField(requester, "maxRecentQuestions", 30);
+        when(sessionRepository.findRecentSessionIds(eq(1L), eq(11L), any(Pageable.class)))
+            .thenReturn(List.of(9L, 8L));
+        when(questionPoolRepository.findRecentQuestions(eq(List.of(9L, 8L)), any(Pageable.class)))
+            .thenReturn(List.of("이전 질문 A", "이전 질문 B"));
+
+        requester.onSessionCreated(new SessionCreatedEvent(
+            1L, 11L, SessionMode.TECHNICAL, JobCategory.BACKEND, 5, 3, List.of()
+        ));
+
+        ArgumentCaptor<GenerateQuestionsPayload> payloadCaptor =
+            ArgumentCaptor.forClass(GenerateQuestionsPayload.class);
+        verify(publisher).publishToAi(eq("generate.questions"), payloadCaptor.capture(), any(MessageContext.class));
+        assertThat(payloadCaptor.getValue().recentQuestions())
+            .containsExactly("이전 질문 A", "이전 질문 B");
     }
 
     private RabbitMqProperties.RoutingKeyProperties mockRoutingKeys() {
diff --git a/docs/messaging.md b/docs/messaging.md
@@ -226,7 +226,8 @@
     "mode": "TECHNICAL",
     "jobCategory": "BACKEND",
     "documentIds": [42, 17],
-    "maxQuestions": 10
+    "maxQuestions": 10,
+    "recentQuestions": ["이전 면접 질문 텍스트", "..."]
   },
   "context": { "userId": 123, "sessionId": 99 }
 }

Original file line number	Diff line number	Diff line change
`@@ -90,6 +90,7 @@ async def handle(self, message: AbstractIncomingMessage) -> None:`
`90`	`90`	`mode=req.mode,`
`91`	`91`	`max_questions=effective_pool_size,`
`92`	`92`	`context=context_text,`
	`93`	`+ recent_questions=req.recent_questions,`
`93`	`94`	`)`
`94`	`95`
`95`	`96`	`payload = QuestionPoolCallbackPayload(`