Team-StackUp · i3months · Jun 25, 2026 · Jun 25, 2026
diff --git a/ai/src/ai_server/chain/prompts/question_generation.py b/ai/src/ai_server/chain/prompts/question_generation.py
@@ -51,6 +51,9 @@
     "   - TECHNICAL: CS_FUNDAMENTAL + TECH_CHOICE + PROJECT_DEEP_DIVE 중심.\n"
     "   - PERSONALITY: BEHAVIORAL 중심.\n"
     "   - INTEGRATED: 모두 균형.\n"
+    "   - JOB_TAILORED: JD 요구를 검증하는 TECH_CHOICE·PROJECT_DEEP_DIVE 와 지원동기·"
+    "직무적합(컬처핏)을 묻는 BEHAVIORAL 을 균형 있게 섞습니다. JD 요구 대비 갭·지원동기 질문은 "
+    "지원자 자료에 근거가 없을 수 있으니 그 경우 target_evidence 를 비워도 됩니다(JD 인용은 가능).\n"
     "   직군이 여러 개면 각 직군에 질문이 고르게 분배되도록 합니다.\n"
     "   자료가 받쳐주는 만큼만 분배하고, 근거 없는 카테고리를 억지로 채우지 마세요.\n"
     "4. 각 질문에 target_evidence 와 expected_signal 을 채웁니다:\n"

diff --git a/ai/src/ai_server/messaging/consumers/feedback_consumer.py b/ai/src/ai_server/messaging/consumers/feedback_consumer.py
@@ -63,6 +63,7 @@ def __init__(
         job_fit_evaluator: JobFitEvaluator | None = None,
         answer_coach: AnswerCoach | None = None,
         coaching_max_answers: int = 30,
+        coaching_concurrency: int = 5,
     ) -> None:
         self._generator = generator
         self._publisher = publisher
@@ -75,6 +76,7 @@ def __init__(
         self._job_fit_evaluator = job_fit_evaluator
         self._answer_coach = answer_coach
         self._coaching_max_answers = coaching_max_answers
+        self._coaching_concurrency = max(1, coaching_concurrency)
 
     async def handle(self, message: AbstractIncomingMessage) -> None:
         async with message.process(requeue=False):
@@ -135,9 +137,11 @@ async def handle(self, message: AbstractIncomingMessage) -> None:
                     self._coach_answers(req, rag_context),
                 )
             )
-            if self_intro_item is not None:
-                result.panel_breakdown.append(self_intro_item)
-            result.panel_breakdown.extend(job_fit_items)
+            # 빈 평가위원 항목(점수·내용 모두 없음)은 표시하지 않는다 — LLM 부분 응답이 빈 패널로 새는 것 방지.
+            extras = [self_intro_item, *job_fit_items]
+            result.panel_breakdown.extend(
+                e for e in extras if e is not None and _panel_has_content(e)
+            )
 
             payload = FeedbackCallbackPayload(
                 session_id=req.session_id,
@@ -255,20 +259,23 @@ async def _coach_answers(
             )
             pairs = pairs[: self._coaching_max_answers]
         target_role = _coaching_target_role(req)
+        # 게이트웨이 429/과부하 방지 — 답변별 코칭 호출 동시성을 제한한다.
+        sem = asyncio.Semaphore(self._coaching_concurrency)
 
         async def _one(
             question: FeedbackMessageItem, answer: FeedbackMessageItem
         ) -> AnswerCoachingItem | None:
             try:
-                res = await self._answer_coach.coach(
-                    job_category=req.job_category,
-                    mode=req.mode,
-                    target_role=target_role,
-                    question=question.content,
-                    expected_signal=question.expected_signal or "",
-                    answer=answer.content,
-                    rag_context=rag_context,
-                )
+                async with sem:
+                    res = await self._answer_coach.coach(
+                        job_category=req.job_category,
+                        mode=req.mode,
+                        target_role=target_role,
+                        question=question.content,
+                        expected_signal=question.expected_signal or "",
+                        answer=answer.content,
+                        rag_context=rag_context,
+                    )
             except Exception as exc:  # noqa: BLE001
                 log.warning(
                     "feedback.coaching.failed",
@@ -346,6 +353,16 @@ def _coaching_target_role(req: GenerateFeedbackRequest) -> str:
     return f"{head}채용공고(JD) 발췌: {jd[:800]}"
 
 
+def _panel_has_content(item: PanelBreakdownItem) -> bool:
+    """점수도 서술도 없는 빈 평가위원 항목인지. 빈 항목은 표시에서 제외한다."""
+    return (
+        item.score is not None
+        or bool((item.detail or "").strip())
+        or bool((item.strength or "").strip())
+        or bool((item.weakness or "").strip())
+    )
+
+
 def _to_panel_item(
     label: str, dimension: str, ev: EvaluatorResult
 ) -> PanelBreakdownItem:

diff --git a/ai/src/ai_server/messaging/consumers/questions_consumer.py b/ai/src/ai_server/messaging/consumers/questions_consumer.py
@@ -170,6 +170,15 @@ def _build_initial_rag_query(req: GenerateQuestionsRequest) -> str:
         f"mode: {req.mode}",
         f"job categories: {', '.join(req.job_categories)}",
     ]
+    # 자기소개·JD 를 쿼리에 포함해 검색이 지원자 발화/직무 요구로 steering 되게 한다.
+    intro = (req.self_intro_answer or "").strip()
+    if intro:
+        parts.append(f"self-introduction: {intro[:600]}")
+    jd = (req.target_job_description or "").strip()
+    if jd:
+        company = (req.target_company_name or "").strip()
+        head = f"target role at {company}" if company else "target role (JD)"
+        parts.append(f"{head}: {jd[:800]}")
     for d in req.documents:
         doc_parts = [f"document #{d.document_id} {d.source_type}"]
         if d.summary:

diff --git a/ai/tests/test_feedback_consumer.py b/ai/tests/test_feedback_consumer.py
@@ -595,6 +595,37 @@ async def test_consumer_appends_job_fit_panel_item():
     assert und_items[0].score == 60.0
 
 
+@pytest.mark.asyncio
+async def test_consumer_drops_empty_job_fit_axis():
+    # understanding 축이 빈(점수·내용 없음) 경우 그 패널 항목은 표시에서 제외된다.
+    generator = _generator()
+    publisher = MagicMock()
+    publisher.publish = AsyncMock()
+    evaluator = MagicMock()
+    evaluator.evaluate = AsyncMock(
+        return_value=JobFitResult(
+            fit=EvaluatorResult(score=72.0, detail="결제 경험 부합"),
+            understanding=EvaluatorResult(),  # 빈 축
+        )
+    )
+
+    consumer = FeedbackConsumer(
+        generator=generator,
+        publisher=publisher,
+        idempotency=LruIdempotencyStore(max_size=10),
+        callback_routing_key="callback.feedback",
+        core_client=MagicMock(),
+        embedder=None,
+        job_fit_evaluator=evaluator,
+    )
+    await consumer.handle(_StubMessage(_job_tailored_envelope()))
+
+    payload: FeedbackCallbackPayload = publisher.publish.await_args.kwargs["payload"]
+    labels = [b.evaluator for b in payload.panel_breakdown]
+    assert "직무 적합도" in labels
+    assert "직무 이해도" not in labels  # 빈 축은 제외
+
+
 @pytest.mark.asyncio
 async def test_consumer_skips_job_fit_when_not_job_tailored():
     generator = _generator()

diff --git a/ai/tests/test_questions_consumer.py b/ai/tests/test_questions_consumer.py
@@ -141,7 +141,9 @@ async def test_consumer_forwards_self_intro_answer_to_generator():
     await consumer.handle(_StubMessage(body))
 
     call = generator.generate.await_args
-    assert call.kwargs["self_introduction"] == "결제 시스템을 설계한 백엔드 3년차입니다."
+    assert (
+        call.kwargs["self_introduction"] == "결제 시스템을 설계한 백엔드 3년차입니다."
+    )
 
 
 @pytest.mark.asyncio
@@ -386,6 +388,27 @@ async def test_consumer_falls_back_to_document_context_when_rag_fails():
     assert "Retrieved document chunks" not in context
 
 
+def test_build_initial_rag_query_includes_self_intro_and_jd():
+    from ai_server.messaging.consumers.questions_consumer import (
+        _build_initial_rag_query,
+    )
+    from ai_server.model.messages.questions import GenerateQuestionsRequest
+
+    req = GenerateQuestionsRequest(
+        session_id=1,
+        mode="JOB_TAILORED",
+        job_categories=["BACKEND"],
+        documents=[],
+        self_intro_answer="결제 시스템을 만든 백엔드 3년차",
+        target_company_name="토스",
+        target_job_description="대용량 결제 백엔드, Kotlin/Spring",
+    )
+    q = _build_initial_rag_query(req)
+    assert "결제 시스템을 만든 백엔드" in q  # 자기소개가 쿼리에 포함
+    assert "대용량 결제 백엔드" in q  # JD 가 쿼리에 포함
+    assert "토스" in q
+
+
 def test_build_context_handles_empty_documents():
     assert _build_context([]) == "(no documents)"
 

diff --git a/backend/src/main/java/com/stackup/stackup/session/application/FeedbackCallbackService.java b/backend/src/main/java/com/stackup/stackup/session/application/FeedbackCallbackService.java
@@ -12,6 +12,7 @@
 import com.stackup.stackup.session.application.dto.FeedbackCallbackPayload;
 import com.stackup.stackup.session.domain.InterviewMessage;
 import com.stackup.stackup.session.domain.InterviewMessageRepository;
+import com.stackup.stackup.session.domain.MessageRole;
 import com.stackup.stackup.session.domain.InterviewSession;
 import com.stackup.stackup.session.domain.InterviewSessionRepository;
 import com.stackup.stackup.session.domain.SessionFeedback;
@@ -112,7 +113,10 @@ private void applyAnswerCoaching(Long sessionId, java.util.List<AnswerCoachingIt
                 continue;
             }
             InterviewMessage message = messageRepository.findById(item.messageId()).orElse(null);
-            if (message == null || !sessionId.equals(message.getSession().getId())) {
+            // 답변(INTERVIEWEE) 메시지에만 기록 — 잘못된 messageId 로 질문에 복기가 달리는 것 방지.
+            if (message == null
+                || !sessionId.equals(message.getSession().getId())
+                || message.getRole() != MessageRole.INTERVIEWEE) {
                 continue;
             }
             message.recordCoaching(item.modelAnswer(), item.answerRewrite(), item.coachingComment());

diff --git a/backend/src/main/java/com/stackup/stackup/session/application/QuestionsCallbackService.java b/backend/src/main/java/com/stackup/stackup/session/application/QuestionsCallbackService.java
@@ -14,6 +14,7 @@
 import com.stackup.stackup.session.domain.InterviewMessageRepository;
 import com.stackup.stackup.session.domain.InterviewSession;
 import com.stackup.stackup.session.domain.InterviewSessionRepository;
+import com.stackup.stackup.session.domain.MessageRole;
 import com.stackup.stackup.session.domain.SessionQuestionPool;
 import com.stackup.stackup.session.domain.SessionQuestionPoolRepository;
 import com.stackup.stackup.session.domain.SessionStatus;
@@ -236,11 +237,14 @@ private void recordAnswerEvaluation(QuestionsCallbackPayload payload) {
         if (eval == null || payload.answerMessageId() == null) {
             return;
         }
-        messageRepository.findById(payload.answerMessageId()).ifPresent(answer -> {
-            answer.recordAnswerEvaluation(
-                eval.specificity(), eval.logic(), eval.structure(), eval.correctness());
-            messageRepository.save(answer);
-        });
+        messageRepository.findById(payload.answerMessageId())
+            // 답변(INTERVIEWEE) 메시지에만 평가 기록 — 잘못된 messageId 로 질문에 점수가 달리는 것 방지.
+            .filter(answer -> answer.getRole() == MessageRole.INTERVIEWEE)
+            .ifPresent(answer -> {
+                answer.recordAnswerEvaluation(
+                    eval.specificity(), eval.logic(), eval.structure(), eval.correctness());
+                messageRepository.save(answer);
+            });
     }
 
     private boolean isProcessed(String messageId) {

diff --git a/backend/src/main/java/com/stackup/stackup/session/presentation/dto/SessionCreateRequest.java b/backend/src/main/java/com/stackup/stackup/session/presentation/dto/SessionCreateRequest.java
@@ -13,7 +13,7 @@
 public record SessionCreateRequest(
     @Size(max = 200) String title,
     @Size(max = 4000) String memo,
-    // mode: TECHNICAL/PERSONALITY/INTEGRATED 중 하나를 선택한다.
+    // mode: TECHNICAL/PERSONALITY/INTEGRATED/JOB_TAILORED 중 하나. JOB_TAILORED 는 아래 회사명·JD 필수.
     @NotNull SessionMode mode,
     // 직군 다중 선택. 최소 1개. 첫 항목이 대표 직군이 된다.
     @NotEmpty List<JobCategory> jobCategories,

diff --git a/frontend/src/domain/session/lib/turn.ts b/frontend/src/domain/session/lib/turn.ts
@@ -43,7 +43,11 @@ export function canSubmitAnswer(session: Session, messages: Message[]): boolean
 }
 
 export function sessionProgress(session: Session): { current: number; max: number; ratio: number } {
-  const max = session.maxQuestions ?? 0
+  // 실제 면접은 일반질문 수(자기소개 포함 = generalQuestionCount)에서 끝나거나 maxQuestions 상한에서
+  // 끝난다 — 둘 중 작은 값이 진짜 목표. maxQuestions(상한)로만 나누면 진행률이 과소 표시된다.
+  const cap = session.maxQuestions ?? 0
+  const general = session.generalQuestionCount ?? cap
+  const max = cap > 0 ? Math.min(general || cap, cap) : general
   const current = session.totalQuestionCount ?? 0
   const ratio = max > 0 ? Math.min(current / max, 1) : 0
   return { current, max, ratio }

diff --git a/frontend/src/features/history/ui/ScoreTrend.tsx b/frontend/src/features/history/ui/ScoreTrend.tsx
@@ -45,12 +45,19 @@ export function ScoreTrend({ stats }: { stats: UserStats }) {
           : null,
       )
       .filter((p): p is { x: number; y: number } => p !== null)
-    const vals = sessions
-      .map((s) => s[m.key])
-      .filter((v): v is number => typeof v === 'number')
-    const latest = vals.length ? Math.round(vals[vals.length - 1]) : null
+    // 최신 점수 + 지난번(바로 직전 세션) 대비 델타. sparse 지표에서 비인접 세션끼리 비교하지 않는다.
+    const seq = sessions.map((s) => (typeof s[m.key] === 'number' ? (s[m.key] as number) : null))
+    let latestIdx = -1
+    for (let i = seq.length - 1; i >= 0; i--) {
+      if (seq[i] !== null) {
+        latestIdx = i
+        break
+      }
+    }
+    const latest = latestIdx >= 0 ? Math.round(seq[latestIdx] as number) : null
+    const prev = latestIdx > 0 ? seq[latestIdx - 1] : null
     const delta =
-      vals.length >= 2 ? Math.round(vals[vals.length - 1] - vals[vals.length - 2]) : null
+      latest !== null && prev !== null ? Math.round((seq[latestIdx] as number) - prev) : null
     return { ...m, pts, latest, delta }
   })
 

diff --git a/frontend/src/features/interview/ui/live/AnswerCoachingAccordion.tsx b/frontend/src/features/interview/ui/live/AnswerCoachingAccordion.tsx
@@ -77,15 +77,18 @@ export function AnswerCoachingAccordion({ message }: { message: Message }) {
             </div>
           )}
           {evals.length > 0 && (
-            <div className="flex flex-wrap gap-1.5">
-              {evals.map((e) => (
-                <span
-                  key={e}
-                  className="rounded-pill bg-surface-raised px-2 py-0.5 text-caption text-fg-muted"
-                >
-                  {e}
-                </span>
-              ))}
+            <div className="flex flex-col gap-1">
+              <span className="text-caption text-fg-subtle">답변 평가</span>
+              <div className="flex flex-wrap gap-1.5">
+                {evals.map((e) => (
+                  <span
+                    key={e}
+                    className="rounded-pill bg-surface-raised px-2 py-0.5 text-caption text-fg-muted"
+                  >
+                    {e}
+                  </span>
+                ))}
+              </div>
             </div>
           )}
           {message.coachingComment && (

diff --git a/frontend/src/features/interview/ui/setup/InterviewSetupForm.tsx b/frontend/src/features/interview/ui/setup/InterviewSetupForm.tsx
@@ -85,7 +85,17 @@ export function InterviewSetupForm({
       </section>
       <section className="flex flex-col gap-2">
         <h2 className="text-h6 text-fg">면접 모드</h2>
-        <ModeSelector value={mode} onChange={setMode} />
+        <ModeSelector
+          value={mode}
+          onChange={(m) => {
+            setMode(m)
+            // 직무 맞춤이 아니면 회사명·JD 입력값을 비운다(전송엔 이미 미포함이지만 상태도 정리).
+            if (m !== 'JOB_TAILORED') {
+              setCompanyName('')
+              setJobDescription('')
+            }
+          }}
+        />
       </section>
       {isJobTailored && (
         <section className="flex flex-col gap-3">
@@ -176,9 +186,16 @@ export function InterviewSetupForm({
         <h2 className="text-h6 text-fg">참고 문서 (선택)</h2>
         <ContextDocumentPicker documents={documents} selected={selected} onToggle={toggle} />
       </section>
-      <Button type="submit" size="lg" loading={isSubmitting} disabled={!valid}>
-        면접 생성
-      </Button>
+      <div className="flex flex-col gap-2">
+        {isJobTailored && jobDescription.trim().length === 0 && (
+          <p className="text-caption text-warning-700">
+            채용공고(JD)를 입력해야 직무 맞춤 면접을 생성할 수 있어요.
+          </p>
+        )}
+        <Button type="submit" size="lg" loading={isSubmitting} disabled={!valid}>
+          면접 생성
+        </Button>
+      </div>
     </form>
   )
 }