#233까지 반영해 배포

src/main/java/com/devkor/ifive/nadab/domain/askchat/application/helper/AskChatAnswerPromptComposer.java

@@ -6,17 +6,27 @@
 import com.devkor.ifive.nadab.domain.askchat.core.dto.AskChatAnswerReferenceDocument;
 import com.devkor.ifive.nadab.domain.askchat.core.properties.AskChatAnswerProperties;
 import com.devkor.ifive.nadab.global.core.prompt.askchat.AskChatAnswerPromptLoader;
+import com.devkor.ifive.nadab.global.core.response.ErrorCode;
+import com.devkor.ifive.nadab.global.exception.ai.AiServiceException;
 import lombok.RequiredArgsConstructor;
 import org.springframework.stereotype.Component;
 
 import java.util.List;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 
 @Component
 @RequiredArgsConstructor
 public class AskChatAnswerPromptComposer implements AskChatAnswerPromptAugmenter {
 
     private static final String NO_REFERENCE_DOCUMENT = "검색된 사용자 기록이 없습니다.";
     private static final String NO_RECENT_MESSAGE = "최근 대화가 없습니다.";
+    private static final Pattern PROMPT_VERSION_SECTION_PATTERN = Pattern.compile(
+            "(?m)^\\[프롬프트 버전]\\R\\{promptVersion}(?:\\R){1,2}"
+    );
+    private static final Pattern TEMPLATE_VARIABLE_PATTERN = Pattern.compile(
+            "\\{([A-Za-z][A-Za-z0-9_]*)}"
+    );
 
     private final AskChatAnswerProperties properties;
     private final AskChatAnswerPromptLoader promptLoader;
@@ -34,12 +44,33 @@ private String systemPrompt() {
     }
 
     private String userPrompt(AskChatAnswerPromptContext context) {
-        return promptLoader.loadUserPrompt()
-                .replace("{promptVersion}", String.valueOf(properties.getPromptVersion()))
-                .replace("{question}", context.question())
-                .replace("{recentMessages}", formatRecentMessages(context.recentMessages()))
-                .replace("{referenceDocuments}", formatReferenceDocuments(context.referenceDocuments()))
-                .replace("{followUpQuestionCount}", String.valueOf(properties.getFollowUpQuestionCount()));
+        String template = PROMPT_VERSION_SECTION_PATTERN
+                .matcher(promptLoader.loadUserPrompt())
+                .replaceFirst("");
+
+        return renderTemplate(template, context);
+    }
+
+    private String renderTemplate(String template, AskChatAnswerPromptContext context) {
+        String question = escapePromptData(context.question());
+        String recentMessages = formatRecentMessages(context.recentMessages());
+        String referenceDocuments = formatReferenceDocuments(context.referenceDocuments());
+        String followUpQuestionCount = String.valueOf(properties.getFollowUpQuestionCount());
+
+        Matcher matcher = TEMPLATE_VARIABLE_PATTERN.matcher(template);
+        StringBuilder builder = new StringBuilder();
+        while (matcher.find()) {
+            String replacement = switch (matcher.group(1)) {
+                case "question" -> question;
+                case "recentMessages" -> recentMessages;
+                case "referenceDocuments" -> referenceDocuments;
+                case "followUpQuestionCount" -> followUpQuestionCount;
+                default -> throw new AiServiceException(ErrorCode.PROMPT_ASK_CHAT_VARIABLE_UNSUPPORTED);
+            };
+            matcher.appendReplacement(builder, Matcher.quoteReplacement(replacement));
+        }
+        matcher.appendTail(builder);
+        return builder.toString();
     }
 
     private String formatRecentMessages(List<AskChatAnswerConversationMessage> messages) {
@@ -54,7 +85,7 @@ private String formatRecentMessages(List<AskChatAnswerConversationMessage> messa
                     .append(". ")
                     .append(message.role())
                     .append(": ")
-                    .append(message.content())
+                    .append(escapePromptData(message.content()))
                     .append(System.lineSeparator());
         }
         return builder.toString().trim();
@@ -68,19 +99,25 @@ private String formatReferenceDocuments(List<AskChatAnswerReferenceDocument> doc
         StringBuilder builder = new StringBuilder();
         for (int i = 0; i < documents.size(); i++) {
             AskChatAnswerReferenceDocument document = documents.get(i);
-            builder.append(i + 1)
-                    .append(". documentId=")
-                    .append(document.documentId())
-                    .append(", sourceType=")
-                    .append(document.sourceType())
-                    .append(", interestCode=")
-                    .append(document.interestCode())
-                    .append(", distance=")
-                    .append(document.distance())
+            builder.append("[기록 ")
+                    .append(i + 1)
+                    .append("]")
+                    .append(System.lineSeparator())
+                    .append(escapePromptData(document.content()))
                     .append(System.lineSeparator())
-                    .append(document.content())
                     .append(System.lineSeparator());
         }
         return builder.toString().trim();
     }
+
+    private String escapePromptData(String value) {
+        if (value == null) {
+            return "";
+        }
+
+        return value
+                .replace("&", "&amp;")
+                .replace("<", "&lt;")
+                .replace(">", "&gt;");
+    }
 }

src/main/java/com/devkor/ifive/nadab/domain/askchat/core/properties/AskChatAnswerProperties.java

@@ -25,6 +25,9 @@ public class AskChatAnswerProperties {
     @NotBlank
     private String model = "gpt-5.6-luna";
 
+    @NotBlank
+    private String reasoningEffort = "low";
+
     @DecimalMin("0.0")
     @DecimalMax("2.0")
     private double temperature = 1.0;
@@ -37,7 +40,4 @@ public class AskChatAnswerProperties {
 
     @Min(0)
     private int followUpQuestionCount = 2;
-
-    @Min(1)
-    private int promptVersion = 2;
 }

src/main/java/com/devkor/ifive/nadab/domain/askchat/infra/AskChatAnswerLlmClient.java

@@ -28,6 +28,8 @@
 @RequiredArgsConstructor
 public class AskChatAnswerLlmClient {
 
+    private static final int MAX_FOLLOW_UP_QUESTION_LENGTH = 30;
+
     /*
      * Keep prompt augmentation behind this boundary while evidence documents are stored in
      * ask_chat_message_references. A future Spring AI Advisor implementation can replace this
@@ -75,6 +77,7 @@ public AskChatAnswerGenerationResult generate(AskChatAnswerPromptContext context
     private OpenAiChatOptions options() {
         return OpenAiChatOptions.builder()
                 .model(properties.getModel())
+                .reasoningEffort(properties.getReasoningEffort())
                 .temperature(properties.getTemperature())
                 .maxCompletionTokens(properties.getMaxTokens())
                 .build();
@@ -93,17 +96,33 @@ private void validateAnswer(AskChatGeneratedAnswer answer) {
             throw new AiResponseParseException(ErrorCode.AI_RESPONSE_FORMAT_INVALID);
         }
 
+        if (containsUnsupportedScript(answer.answer())) {
+            throw new AiResponseParseException(ErrorCode.AI_RESPONSE_UNSUPPORTED_SCRIPT);
+        }
+
         if (answer.followUpQuestions().size() > properties.getFollowUpQuestionCount()) {
             throw new AiResponseParseException(ErrorCode.AI_RESPONSE_FORMAT_INVALID);
         }
 
         for (String followUpQuestion : answer.followUpQuestions()) {
-            if (isBlank(followUpQuestion)) {
+            if (isBlank(followUpQuestion)
+                    || followUpQuestion.codePointCount(0, followUpQuestion.length()) > MAX_FOLLOW_UP_QUESTION_LENGTH) {
                 throw new AiResponseParseException(ErrorCode.AI_RESPONSE_FORMAT_INVALID);
             }
+
+            if (containsUnsupportedScript(followUpQuestion)) {
+                throw new AiResponseParseException(ErrorCode.AI_RESPONSE_UNSUPPORTED_SCRIPT);
+            }
         }
     }
 
+    private boolean containsUnsupportedScript(String value) {
+        return value.codePoints().anyMatch(codePoint -> switch (Character.UnicodeScript.of(codePoint)) {
+            case COMMON, INHERITED, HANGUL, LATIN -> false;
+            default -> true;
+        });
+    }
+
     private List<Long> referenceDocumentIds(AskChatAnswerPromptContext context) {
         return context.referenceDocuments().stream()
                 .map(AskChatAnswerReferenceDocument::documentId)

src/main/java/com/devkor/ifive/nadab/domain/dailyreport/application/DailyReportService.java

@@ -5,11 +5,13 @@
 import com.devkor.ifive.nadab.domain.dailyreport.api.dto.response.CreateAnswerImageUploadUrlResponse;
 import com.devkor.ifive.nadab.domain.dailyreport.api.dto.response.CreateDailyReportResponse;
 import com.devkor.ifive.nadab.domain.dailyreport.api.dto.response.ImageStatusResponse;
+import com.devkor.ifive.nadab.domain.dailyreport.application.helper.DailyReportModelSelector;
 import com.devkor.ifive.nadab.domain.dailyreport.core.dto.ConfirmDailyAndRewardDto;
 import com.devkor.ifive.nadab.domain.dailyreport.core.dto.PrepareDailyResultDto;
 import com.devkor.ifive.nadab.domain.dailyreport.core.dto.AiDailyReportResultDto;
 import com.devkor.ifive.nadab.domain.dailyreport.core.entity.AnswerEntry;
 import com.devkor.ifive.nadab.domain.dailyreport.core.entity.ImageStatus;
+import com.devkor.ifive.nadab.domain.dailyreport.core.properties.DailyReportLlmProperties.ModelCandidate;
 import com.devkor.ifive.nadab.domain.dailyreport.infra.DailyReportLlmClient;
 import com.devkor.ifive.nadab.domain.question.core.entity.DailyQuestion;
 import com.devkor.ifive.nadab.domain.question.core.entity.UserDailyQuestion;
@@ -48,6 +50,7 @@ public class DailyReportService {
     private final DailyReportTxService dailyReportTxService;
     private final ProfileImageService profileImageService;
 
+    private final DailyReportModelSelector dailyReportModelSelector;
     private final DailyReportLlmClient dailyReportLlmClient;
     private final ReportGenerationLogRecorder reportGenerationLogRecorder;
 
@@ -88,19 +91,20 @@ public CreateDailyReportResponse generateDailyReport(Long userId, DailyReportReq
         PrepareDailyResultDto prep = dailyReportTxService.prepareDaily(user, question, request.answer(), isDayPassed, request.objectKey());
 
         AnswerEntry answerEntry = prep.entry();
+        ModelCandidate modelCandidate = dailyReportModelSelector.select();
         Long generationLogId = reportGenerationLogRecorder.start(
                 userId,
                 ReportGenerationType.DAILY,
                 prep.reportId(),
                 ReportGenerationStep.DAILY_GENERATE,
                 LlmProvider.OPENAI,
-                dailyReportLlmClient.model()
+                modelCandidate.getModel()
         );
 
         AiDailyReportResultDto dto;
         try {
             LlmGenerationResult<AiDailyReportResultDto> generationResult =
-                    dailyReportLlmClient.generate(question.getQuestionText(), answerEntry);
+                    dailyReportLlmClient.generate(question.getQuestionText(), answerEntry, modelCandidate);
             dto = generationResult.content();
             LlmTokenUsage tokenUsage = generationResult.tokenUsage();
             reportGenerationLogRecorder.recordTokenUsage(

src/main/java/com/devkor/ifive/nadab/domain/dailyreport/application/helper/DailyReportModelSelector.java

@@ -0,0 +1,45 @@
+package com.devkor.ifive.nadab.domain.dailyreport.application.helper;
+
+import com.devkor.ifive.nadab.domain.dailyreport.core.properties.DailyReportLlmProperties;
+import com.devkor.ifive.nadab.domain.dailyreport.core.properties.DailyReportLlmProperties.ModelCandidate;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.stereotype.Component;
+
+import java.util.concurrent.ThreadLocalRandom;
+import java.util.function.IntUnaryOperator;
+
+@Component
+public class DailyReportModelSelector {
+
+    private static final int TOTAL_WEIGHT = 100;
+
+    private final DailyReportLlmProperties properties;
+    private final IntUnaryOperator randomValueGenerator;
+
+    @Autowired
+    public DailyReportModelSelector(DailyReportLlmProperties properties) {
+        this(properties, bound -> ThreadLocalRandom.current().nextInt(bound));
+    }
+
+    DailyReportModelSelector(
+            DailyReportLlmProperties properties,
+            IntUnaryOperator randomValueGenerator
+    ) {
+        this.properties = properties;
+        this.randomValueGenerator = randomValueGenerator;
+    }
+
+    public ModelCandidate select() {
+        int randomValue = randomValueGenerator.applyAsInt(TOTAL_WEIGHT);
+        int cumulativeWeight = 0;
+
+        for (ModelCandidate candidate : properties.getCandidates()) {
+            cumulativeWeight += candidate.getWeight();
+            if (randomValue < cumulativeWeight) {
+                return candidate;
+            }
+        }
+
+        throw new IllegalStateException("Failed to select a DailyReport LLM model candidate");
+    }
+}

src/main/java/com/devkor/ifive/nadab/domain/dailyreport/core/properties/DailyReportLlmProperties.java

@@ -1,37 +1,86 @@
 package com.devkor.ifive.nadab.domain.dailyreport.core.properties;
 
+import jakarta.validation.Valid;
+import jakarta.validation.constraints.AssertTrue;
 import jakarta.validation.constraints.DecimalMax;
 import jakarta.validation.constraints.DecimalMin;
+import jakarta.validation.constraints.Max;
 import jakarta.validation.constraints.Min;
 import jakarta.validation.constraints.NotBlank;
+import jakarta.validation.constraints.NotEmpty;
 import jakarta.validation.constraints.NotNull;
 import lombok.Getter;
 import lombok.Setter;
 import org.springframework.boot.context.properties.ConfigurationProperties;
 import org.springframework.stereotype.Component;
 import org.springframework.validation.annotation.Validated;
 
+import java.util.ArrayList;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Objects;
+
 @Component
 @Getter
 @Setter
 @Validated
 @ConfigurationProperties(prefix = "daily-report.llm")
 public class DailyReportLlmProperties {
 
-    @NotBlank
-    private String model = "gpt-4o-mini";
+    @Valid
+    @NotEmpty
+    private List<@NotNull ModelCandidate> candidates = new ArrayList<>();
+
+    @AssertTrue(message = "daily-report.llm.candidates weights must total 100")
+    public boolean isCandidateWeightTotalValid() {
+        if (candidates == null || candidates.isEmpty()) {
+            return true;
+        }
+
+        return candidates.stream()
+                .filter(Objects::nonNull)
+                .mapToInt(ModelCandidate::getWeight)
+                .sum() == 100;
+    }
+
+    @AssertTrue(message = "daily-report.llm.candidates models must be unique")
+    public boolean isCandidateModelUnique() {
+        if (candidates == null || candidates.isEmpty()) {
+            return true;
+        }
+
+        List<String> models = candidates.stream()
+                .filter(Objects::nonNull)
+                .map(ModelCandidate::getModel)
+                .filter(Objects::nonNull)
+                .toList();
 
-    @DecimalMin("0.0")
-    @DecimalMax("2.0")
-    private double temperature = 0.3;
+        return new HashSet<>(models).size() == models.size();
+    }
+
+    @Getter
+    @Setter
+    public static class ModelCandidate {
+
+        @NotBlank
+        private String model;
 
-    @Min(1)
-    private int maxOutputTokens = 512;
+        @Min(1)
+        @Max(100)
+        private int weight;
 
-    @NotNull
-    private TokenLimitParameter tokenLimitParameter = TokenLimitParameter.MAX_TOKENS;
+        @DecimalMin("0.0")
+        @DecimalMax("2.0")
+        private double temperature;
 
-    private String reasoningEffort;
+        @Min(1)
+        private int maxOutputTokens;
+
+        @NotNull
+        private TokenLimitParameter tokenLimitParameter;
+
+        private String reasoningEffort;
+    }
 
     public enum TokenLimitParameter {
         MAX_TOKENS,