Develop by terisuke · Pull Request #32 · terisuke/engineer-cafe-navigator

terisuke · 2025-06-30T13:44:10Z

Summary by CodeRabbit

新機能
- 日本語音声認識の精度向上のためのSTT自動補正システムを追加
- 会話メモリシステムが強化され、セッションIDやリクエストタイプの追跡に対応
- 回答精度向上機能を実装し、特定の質問には簡潔な1～2文で返答
- モバイル対応の音声再生システムをWeb Audio APIベースで刷新し、統合AudioPlaybackServiceを導入
- 管理用ダッシュボードやヘルスチェックAPI、運用監視・アラート機能を追加
- 複数エージェント間で共有可能なSharedMemoryServiceを導入し、会話文脈の一元管理を実現
改善
- マルチリンガルRAG検索や知識ベース自動更新（CRON連携）を拡充
- メモリシステムや音声再生のパフォーマンス最適化（リップシンク高速化、TTL管理強化など）
- 回答文から不要なタグやラベルを除去し、より自然で焦点を絞った応答を実現
- STT補正を会話前処理に組み込み、検索・質問分類の精度向上
- 会話の文脈理解とフォローアップ対応を強化し、明確な質問タイプに基づく応答を実装
ドキュメント
- 各種README・API仕様・開発/運用手順を大幅に拡充・最新化
- 新機能や運用監視、トラブルシューティング手順を詳細に追記
- 新規ブログ記事を追加し、会話文脈理解技術を解説
テスト
- STT補正・メモリシステムの動作検証用テストスクリプトを追加
その他
- STT補正ログ用テーブルと集計ビューをDBに新設し、バッチ処理でログ収集・分析を実施
- 不要となった古いテスト・レポートドキュメントを削除
- 開発用依存パッケージ「node-fetch」を追加

When users ask about specific entities like Saino Cafe, the RAG search now prioritizes similarity scores over importance levels to ensure the most relevant results appear first. This fixes the issue where Engineer Cafe results (0.48 similarity) were ranking higher than Saino Cafe results (0.79 similarity) for Saino-specific queries. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

- Remove square bracket tags [Tag Name] from RAG search results - Add explicit instructions to prompts to avoid including tags in responses - Post-process responses to remove any leaked tags - Add response length check for specific requests (max 150 chars) - Extract key information if response is too long for specific requests - Ensure operating hours queries return only the requested information - Clean up both single-language and multi-language search formatting This fixes the issue where the system was returning entire knowledge base entries with tags instead of just the specific information requested (e.g., operating hours). 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

- Add sessionId parameter to EnhancedQAAgent.answerQuestion method - Pass sessionId from API route to agent for proper memory association - Store sessionId with each message in SimplifiedMemorySystem - Add comprehensive debug logging to track memory storage and retrieval - Ensure requestType metadata is preserved in conversation memory This fixes the issue where the agent couldn't remember previous questions when users provided clarifying responses (e.g., "才能の方で" after asking about cafe hours). 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

## STT Correction System - Add automatic speech-to-text correction for common misrecognitions - Implement context-aware corrections for "カフェ" vs "壁" confusion - Add monitoring system to track correction patterns - Include test scripts and database migration ## Documentation Updates - Update README.md with latest features including STT corrections, monitoring, and CRON jobs - Enhance CLAUDE.md with SimplifiedMemorySystem improvements and recent fixes - Update all docs/ files with current implementation status - Add production monitoring and automated update documentation ## Code Improvements - Enhance memory system with sessionId tracking - Fix conversation context retrieval for follow-up questions - Add specific request type extraction and tracking - Improve response precision for contextual questions ## Cleanup - Remove obsolete test result files - Remove historical RAG fix documentation - Clean up outdated implementation logs This comprehensive update brings all documentation in line with the current state of the project and adds critical STT correction functionality to improve user experience. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

- Fix date in README.md from 2025/01/30 to 2025/06/30 - Update README-EN.md with all latest features: - STT correction system for Japanese misrecognition - Enhanced memory system with SimplifiedMemorySystem - Response precision improvements - Production monitoring features - CRON job documentation - Complete audio system refactoring details - All new development commands - Updated project structure - STT misrecognition troubleshooting - Updated performance metrics - Ensure both Japanese and English documentation are in sync 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

- Fix module not found error by updating import from './supabase-client' to './supabase' - Resolves build error preventing API routes from loading 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

vercel · 2025-06-30T13:44:14Z

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
engineer-cafe-navigator	❌ Failed (Inspect)			Jul 1, 2025 7:07am

coderabbitai · 2025-06-30T13:44:17Z

Walkthrough

この変更は、音声認識（STT）補正システムの導入・強化、会話メモリシステムのセッションID管理やリクエストタイプ判定の改善、AIエージェントの回答精度向上、監視・分析基盤の拡充、関連ドキュメント・テスト・スクリプトの追加・修正を含む大規模なアップデートです。コード・ドキュメント・SQL・テスト全体にわたり反映されています。

Changes

ファイル/グループ	変更概要
CLAUDE.md, README.md, README-EN.md, docs/*, docs/STATUS.md, docs/DEPLOYMENT.md, docs/DEVELOPMENT.md	ドキュメントの大幅拡充。STT補正、会話メモリ、監視、API、開発・運用コマンド、トラブルシュート、構成・環境変数、最新機能・構成などを詳細化。
docs/API.md, docs/API-ja.md	APIドキュメント拡張。新エンドポイント（知識検索、監視、管理、CRON、ヘルスチェック）、リクエスト・レスポンス例、認証要件追加。
scripts/create-stt-correction-logs-table.sql	STT補正ログ用テーブル・ビュー・RLSポリシーの新規作成。集計・分析用ビューも追加。
src/lib/stt-correction-monitor.ts	STT補正イベントのバッチ記録・統計取得モジュール新規追加。Supabase連携、プロセス終了時のフラッシュ対応。
src/utils/stt-corrections.ts	STT誤認識パターンの自動補正・信頼度調整・品質分析ユーティリティ新規追加。
src/lib/simplified-memory.ts	メモリメタデータにrequestType追加、addMessage/getRecentMessagesの対応、詳細なロギング追加。
src/mastra/agents/enhanced-qa-agent.ts	STT補正統合、answerQuestionにsessionId追加、リクエストタイプ抽出・プロンプト最適化・応答の1文要約・タグ除去、メモリ保存強化、ロギング強化。
src/mastra/agents/realtime-agent.ts	STT補正ユーティリティの導入、normalizeInputでの補正処理を共通化。
src/mastra/tools/rag-search.ts	特定クエリ時の類似度優先ソート、タイトル除去による出力フォーマット変更。
src/mastra/voice/google-cloud-voice-simple.ts	STT結果に補正処理・信頼度調整を追加、ロギング強化。
src/app/api/qa/route.ts	qaAgent.answerQuestion呼び出しにsessionId引数を追加。
scripts/test-stt-corrections.ts	STT補正ロジックのテストスクリプト新規追加。
src/test/test-memory-storage.ts	メモリシステムのストレージ・取得・統計・セッション要約テストスクリプト新規追加。
package.json	devDependenciesにnode-fetch追加。
src/lib/shared-memory-service.ts	複数エージェント間で共有可能な会話メモリサービスを新規追加。セッションID・言語管理、メッセージ追加・取得、会話状態判定などを提供。
src/mastra/index.ts	エージェント初期化時にsharedMemoryサービスを取得し、modelConfigにsharedMemoryをセット。既存のmemory参照を置換。
src/test/source-prioritization-test-results.md, src/test/test-summary.md, docs/rag-fix-completed.md, docs/rag-improvement-action-summary.md	古いテスト・アクションサマリ・レポートドキュメントの削除。

Sequence Diagram(s)

STT補正システム & QAエージェント連携

sequenceDiagram
    participant User
    participant GoogleCloudVoiceSimple
    participant STT Corrections
    participant EnhancedQAAgent
    participant SimplifiedMemorySystem
    participant Supabase

    User->>GoogleCloudVoiceSimple: 音声入力
    GoogleCloudVoiceSimple->>GoogleCloudVoiceSimple: Google STTで文字起こし
    GoogleCloudVoiceSimple->>STT Corrections: applySttCorrections(transcript)
    STT Corrections-->>GoogleCloudVoiceSimple: 補正済みテキスト・信頼度
    GoogleCloudVoiceSimple->>EnhancedQAAgent: 補正済みテキスト
    EnhancedQAAgent->>EnhancedQAAgent: リクエストタイプ抽出・履歴参照
    EnhancedQAAgent->>SimplifiedMemorySystem: メッセージ保存(sessionId, requestType)
    EnhancedQAAgent->>Supabase: ナレッジ検索・履歴取得
    EnhancedQAAgent->>User: 精密化された回答を返却

STT補正ログ・監視

sequenceDiagram
    participant STT Corrections
    participant SttCorrectionMonitor
    participant Supabase

    STT Corrections->>SttCorrectionMonitor: logCorrection(補正イベント)
    SttCorrectionMonitor->>SttCorrectionMonitor: バッファに蓄積
    Note over SttCorrectionMonitor: 10件または1分ごとに
    SttCorrectionMonitor->>Supabase: stt_correction_logsへバッチ書き込み

Possibly related PRs

#31:
STT補正システムやEnhancedQAAgentの応答精度向上、AudioPlaybackServiceの統合など、本PRと直接的なコード・設計レベルの関連あり。
#25:
3分TTLの短期メモリ導入。会話メモリ強化という観点で本PRと関連するが、本PRはさらにSTT補正や応答精度、監視なども含む拡張。
#15:
EnhancedQAAgentのSTT補正統合やリクエストタイプ判定、応答精度向上に関する機能強化が本PRと関連。両PRとも質問処理ロジックの改善を含む。

Poem

うさぎの耳で聞き分ける、
STTの誤りもすぐ直す。
メモリは賢く会話をつなぎ、
質問の意図もピンポイント！
モニタリングで健康チェック、
みんなの声がもっとクリアに。
ぴょんぴょん跳ねて、今日も進化！ 🐇✨

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 ESLint

If the error stems from missing dependencies, add them to the package.json file. For unrecoverable errors (e.g., due to private dependencies), disable the tool in the CodeRabbit configuration.

src/mastra/agents/enhanced-qa-agent.ts

Oops! Something went wrong! :(

ESLint: 9.27.0

ESLint couldn't find an eslint.config.(js|mjs|cjs) file.

From ESLint v9.0.0, the default configuration file is now eslint.config.js.
If you are using a .eslintrc.* file, please follow the migration guide
to update your configuration file to the new format:

https://eslint.org/docs/latest/use/configure/migration-guide

If you still have problems after following the migration guide, please stop by
https://eslint.org/chat/help to chat with the team.

src/lib/shared-memory-service.ts

Oops! Something went wrong! :(

ESLint: 9.27.0

ESLint couldn't find an eslint.config.(js|mjs|cjs) file.

From ESLint v9.0.0, the default configuration file is now eslint.config.js.
If you are using a .eslintrc.* file, please follow the migration guide
to update your configuration file to the new format:

https://eslint.org/docs/latest/use/configure/migration-guide

If you still have problems after following the migration guide, please stop by
https://eslint.org/chat/help to chat with the team.

src/mastra/agents/realtime-agent.ts

Oops! Something went wrong! :(

ESLint: 9.27.0

ESLint couldn't find an eslint.config.(js|mjs|cjs) file.

From ESLint v9.0.0, the default configuration file is now eslint.config.js.
If you are using a .eslintrc.* file, please follow the migration guide
to update your configuration file to the new format:

https://eslint.org/docs/latest/use/configure/migration-guide

If you still have problems after following the migration guide, please stop by
https://eslint.org/chat/help to chat with the team.

2 others

✨ Finishing Touches

📝 Generate Docstrings

🧪 Generate Unit Tests

Create PR with Unit Tests
Post Copyable Unit Tests in Comment
Commit Unit Tests in branch develop

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

Review comments: Directly reply to a review comment made by CodeRabbit. Example:
- I pushed a fix in commit <commit_id>, please review it.
- Explain this complex logic.
- Open a follow-up GitHub issue for this discussion.
Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
- @coderabbitai explain this code block.
- @coderabbitai modularize this function.
PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
- @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
- @coderabbitai read src/utils.ts and explain its main purpose.
- @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
- @coderabbitai help me debug CodeRabbit configuration file.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

@coderabbitai pause to pause the reviews on a PR.
@coderabbitai resume to resume the paused reviews.
@coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
@coderabbitai full review to do a full review from scratch and review all the files again.
@coderabbitai summary to regenerate the summary of the PR.
@coderabbitai generate docstrings to generate docstrings for this PR.
@coderabbitai generate sequence diagram to generate a sequence diagram of the changes in this PR.
@coderabbitai auto-generate unit tests to generate unit tests for this PR.
@coderabbitai resolve resolve all the CodeRabbit review comments.
@coderabbitai configuration to show the current CodeRabbit configuration for the repository.
@coderabbitai help to get help.

Other keywords and placeholders

Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
Add @coderabbitai anywhere in the PR title to generate the title automatically.

CodeRabbit Configuration File (`.coderabbit.yaml`)

You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
Please see the configuration documentation for more information.
If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Documentation and Community

Visit our Documentation for detailed information on how to use CodeRabbit.
Join our Discord Community to get help, request features, and share feedback.
Follow us on X/Twitter for updates and announcements.

coderabbitai

Actionable comments posted: 7

🧹 Nitpick comments (7)

src/app/api/qa/route.ts (1)

62-62: sessionIdパラメータの追加とswitch文の改善

sessionIdパラメータの追加により、セッション認識機能が強化されました。ただし、静的解析ツールが指摘するswitch文内での変数宣言の問題を修正することをお勧めします。

以下の修正を適用してswitch文の安全性を向上させてください：

      case 'ask_question':
+       {
          if (!question) {
            return NextResponse.json(
              { error: 'Question text required' },
              { status: 400 }
            );
          }

          // Set language if provided
          if (language) {
            await qaAgent.memory.store('language', language);
          }

          const answer = await qaAgent.answerQuestion(question, language, sessionId);
          const category = await qaAgent.categorizeQuestion(question);
          
          // Convert answer to speech
          const voiceService = navigator.getTool('voiceService');
          let audioResponse = null;
          if (voiceService) {
            const currentLanguage = await qaAgent.memory.get('language') || 'ja';
            const audioBuffer = await voiceService.textToSpeech(answer, { language: currentLanguage });
            audioResponse = Buffer.from(audioBuffer).toString('base64');
          }

          // Log the interaction
          const externalTool = navigator.getTool('externalApi');
          if (externalTool && sessionId) {
            await externalTool.execute({
              action: 'logVisitorActivity',
              data: {
                sessionId,
                activity: 'qa_interaction',
                details: {
                  question,
                  answer,
                  category,
                  timestamp: new Date().toISOString(),
                },
              },
            });
          }
          
          return NextResponse.json({
            success: true,
            answer,
            category,
            audioResponse,
            hasAudio: !!audioResponse,
          });
+       }

src/mastra/tools/rag-search.ts (1)

71-84: 特定クエリ検出のハードコード値を設定可能にすることを検討してください

特定クエリの判定条件（"saino", "才能", "併設"）と類似度差の閾値（0.2）がハードコードされています。これらの値を設定可能にすることで、より柔軟な調整が可能になります。

以下のように設定値として外部化することを提案します：

+// Configuration for specific query handling
+const SPECIFIC_QUERY_KEYWORDS = ['saino', '才能', '併設'];
+const SIMILARITY_THRESHOLD_DIFF = 0.2;
+
   // Check if this is a specific query (e.g., Saino Cafe)
-  const isSpecificQuery = query.toLowerCase().includes('saino') || 
-                         query.toLowerCase().includes('才能') ||
-                         query.toLowerCase().includes('併設');
+  const isSpecificQuery = SPECIFIC_QUERY_KEYWORDS.some(keyword => 
+    query.toLowerCase().includes(keyword.toLowerCase()));

   const sorted = results.sort((a,b)=>{
     // For specific queries, prioritize similarity over importance
     if (isSpecificQuery) {
       // If similarity difference is significant (>0.2), use similarity
       const simDiff = b.similarity - a.similarity;
-      if (Math.abs(simDiff) > 0.2) {
+      if (Math.abs(simDiff) > SIMILARITY_THRESHOLD_DIFF) {
         return simDiff;
       }
     }

scripts/test-stt-corrections.ts (1)

38-59: テスト実行とレポート機能の品質向上を提案します

現在のテスト実行ロジックは基本的な検証を行っていますが、以下の改善を検討してください：

 for (const testCase of testCases) {
   const result = applySttCorrections(testCase.input);
   const isCorrect = result === testCase.expected;
+  const executionTime = performance.now();
+  const finalTime = performance.now() - executionTime;
   
   if (isCorrect) {
     passed++;
     console.log(`✅ PASS`);
   } else {
     failed++;
     console.log(`❌ FAIL`);
   }
   
   console.log(`   Input:    "${testCase.input}"`);
   console.log(`   Expected: "${testCase.expected}"`);
   console.log(`   Got:      "${result}"`);
+  console.log(`   Time:     ${finalTime.toFixed(2)}ms`);
   
   if (likelyContainsMisrecognition(testCase.input)) {
     console.log(`   ⚠️  Likely contains misrecognition`);
   }

docs/STATUS.md (1)

61-61: 環境変数リストの文法を修正してください

静的解析ツールが指摘する通り、連続するカンマが存在します。

 ✅ Core functionality:
 - `GOOGLE_CLOUD_PROJECT_ID`, `GOOGLE_CLOUD_CREDENTIALS`, `GOOGLE_GENERATIVE_AI_API_KEY`
-- `NEXT_PUBLIC_SUPABASE_URL`, `NEXT_PUBLIC_SUPABASE_ANON_KEY`, `SUPABASE_SERVICE_ROLE_KEY`
+- `NEXT_PUBLIC_SUPABASE_URL`, `NEXT_PUBLIC_SUPABASE_ANON_KEY`, `SUPABASE_SERVICE_ROLE_KEY`
 - `OPENAI_API_KEY` - For embeddings (1536 dimensions)

src/lib/stt-correction-monitor.ts (1)

19-24: 静的クラス設計の再考を提案

静的解析ツールが指摘している通り、すべてのメンバーが静的なクラスよりも、モジュールレベルの関数を使用する方が適切かもしれません。特にテスタビリティと依存性注入の観点から改善の余地があります。
-export class SttCorrectionMonitor {
-  private static corrections: CorrectionEvent[] = [];
-  private static readonly BATCH_SIZE = 10;
-  private static readonly FLUSH_INTERVAL = 60000;
-  private static flushTimer: NodeJS.Timeout | null = null;
+// モジュールレベルの実装への変更を検討
+let corrections: CorrectionEvent[] = [];
+const BATCH_SIZE = 10;
+const FLUSH_INTERVAL = 60000;
+let flushTimer: NodeJS.Timeout | null = null;

+export const SttCorrectionMonitor = {
+  logCorrection,
+  getStats,
+  shutdown
+} as const;

src/utils/stt-corrections.ts (1)

82-108: 補正ルール適用ロジックの改善を提案します。

現在の実装は基本的には正しいですが、補正の順序や重複適用の防止について改善の余地があります。

 // Apply each correction rule
 for (const rule of CORRECTION_RULES) {
+   // Skip if pattern doesn't match to avoid unnecessary regex operations
+   if (!rule.pattern.test(corrected)) {
+     continue;
+   }
+   
   // Check if context is required and matches
   if (rule.context) {
     if (!rule.context.test(corrected)) {
       continue; // Skip this rule if context doesn't match
     }
   }

-   // Check if the pattern matches
-   if (rule.pattern.test(corrected)) {
     const before = corrected;
     corrected = corrected.replace(rule.pattern, rule.replacement);
     
     if (before !== corrected) {
       corrections.push(rule.description);
+       // Reset pattern to ensure we reprocess the text for other rules
+       rule.pattern.lastIndex = 0;
     }
-   }
 }

docs/API.md (1)

689-689: タイポの修正が必要です。

"embedingModel" は "embeddingModel" であるべきです。
-  "embedingModel": "text-embedding-004",
+  "embeddingModel": "text-embedding-004",

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 29927ae and 370248c.

⛔ Files ignored due to path filters (1)

pnpm-lock.yaml is excluded by !**/pnpm-lock.yaml

📒 Files selected for processing (25)

CLAUDE.md (7 hunks)
README-EN.md (10 hunks)
README.md (6 hunks)
_docs/blog/2025-06-29_AIアシスタントが空気を読めるようになった話〜会話の文脈を理解してピンポイントで答える技術〜.md (1 hunks)
docs/API-ja.md (3 hunks)
docs/API.md (6 hunks)
docs/DEPLOYMENT.md (5 hunks)
docs/DEVELOPMENT.md (6 hunks)
docs/STATUS.md (7 hunks)
docs/rag-fix-completed.md (0 hunks)
docs/rag-improvement-action-summary.md (0 hunks)
package.json (1 hunks)
scripts/create-stt-correction-logs-table.sql (1 hunks)
scripts/test-stt-corrections.ts (1 hunks)
src/app/api/qa/route.ts (1 hunks)
src/lib/simplified-memory.ts (6 hunks)
src/lib/stt-correction-monitor.ts (1 hunks)
src/mastra/agents/enhanced-qa-agent.ts (12 hunks)
src/mastra/agents/realtime-agent.ts (2 hunks)
src/mastra/tools/rag-search.ts (3 hunks)
src/mastra/voice/google-cloud-voice-simple.ts (2 hunks)
src/test/source-prioritization-test-results.md (0 hunks)
src/test/test-memory-storage.ts (1 hunks)
src/test/test-summary.md (0 hunks)
src/utils/stt-corrections.ts (1 hunks)

💤 Files with no reviewable changes (4)

docs/rag-fix-completed.md
src/test/source-prioritization-test-results.md
src/test/test-summary.md
docs/rag-improvement-action-summary.md

🧰 Additional context used

🧠 Learnings (16)

package.json (2)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Always use Tailwind CSS v3.4.17, postcss@8.4.47, and autoprefixer@10.4.20; do not upgrade to Tailwind CSS v4 due to breaking changes and incompatibility.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: PostCSS configuration must use 'tailwindcss: {}' instead of '@tailwindcss/postcss: {}'.

src/mastra/tools/rag-search.ts (1)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

src/test/test-memory-storage.ts (2)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The SimplifiedMemorySystem provides unified, context-aware conversation memory with a 3-minute TTL, agent isolation, and automatic cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

docs/DEVELOPMENT.md (2)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

docs/STATUS.md (7)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Always use Tailwind CSS v3.4.17, postcss@8.4.47, and autoprefixer@10.4.20; do not upgrade to Tailwind CSS v4 due to breaking changes and incompatibility.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: All admin and management interfaces (e.g., knowledge base, categories, metadata templates) are accessible via dedicated API endpoints and require proper authentication.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Monitoring and alerting are implemented via real-time dashboards and webhook endpoints, tracking RAG performance, cache efficiency, API usage, and error rates.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

_docs/blog/2025-06-29_AIアシスタントが空気を読めるようになった話〜会話の文脈を理解してピンポイントで答える技術〜.md (3)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

src/mastra/agents/realtime-agent.ts (1)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

src/lib/simplified-memory.ts (2)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The SimplifiedMemorySystem provides unified, context-aware conversation memory with a 3-minute TTL, agent isolation, and automatic cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

scripts/create-stt-correction-logs-table.sql (1)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: All database tables must have Row Level Security (RLS) enabled, with service role access for secure operations.

README.md (5)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Audio playback must use Web Audio API as primary, with HTMLAudioElement as fallback, and include device-specific optimizations for iOS/Android.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

README-EN.md (10)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Audio playback must use Web Audio API as primary, with HTMLAudioElement as fallback, and include device-specific optimizations for iOS/Android.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Lip-sync analysis uses optimized frequency band analysis (O(n)), not heavy FFT, and includes caching with hybrid memory/localStorage and auto-cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: For iPad/iOS Safari, always initialize AudioContext after explicit user interaction and provide fallback to audio-only mode if lip-sync fails.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The AudioPlaybackService provides a unified API for all audio playback, supporting optional lip-sync and consistent error handling.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The SimplifiedMemorySystem provides unified, context-aware conversation memory with a 3-minute TTL, agent isolation, and automatic cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Monitoring and alerting are implemented via real-time dashboards and webhook endpoints, tracking RAG performance, cache efficiency, API usage, and error rates.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: All admin and management interfaces (e.g., knowledge base, categories, metadata templates) are accessible via dedicated API endpoints and require proper authentication.

docs/DEPLOYMENT.md (5)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Monitoring and alerting are implemented via real-time dashboards and webhook endpoints, tracking RAG performance, cache efficiency, API usage, and error rates.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: For iPad/iOS Safari, always initialize AudioContext after explicit user interaction and provide fallback to audio-only mode if lip-sync fails.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Audio playback must use Web Audio API as primary, with HTMLAudioElement as fallback, and include device-specific optimizations for iOS/Android.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: AudioContext initialization should be deferred until the first user interaction to comply with mobile browser autoplay policies.

docs/API-ja.md (4)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: All admin and management interfaces (e.g., knowledge base, categories, metadata templates) are accessible via dedicated API endpoints and require proper authentication.

docs/API.md (7)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: All admin and management interfaces (e.g., knowledge base, categories, metadata templates) are accessible via dedicated API endpoints and require proper authentication.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Monitoring and alerting are implemented via real-time dashboards and webhook endpoints, tracking RAG performance, cache efficiency, API usage, and error rates.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The SimplifiedMemorySystem provides unified, context-aware conversation memory with a 3-minute TTL, agent isolation, and automatic cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

CLAUDE.md (9)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Always use Tailwind CSS v3.4.17, postcss@8.4.47, and autoprefixer@10.4.20; do not upgrade to Tailwind CSS v4 due to breaking changes and incompatibility.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: PostCSS configuration must use 'tailwindcss: {}' instead of '@tailwindcss/postcss: {}'.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Automated CRON jobs update the knowledge base and slides every 6 hours, secured by a CRON_SECRET, with error recovery and retry logic.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The application architecture is multi-layered: frontend (Next.js/React), AI agent (Mastra), audio (Web Audio API), and data (Supabase/PostgreSQL with pgvector).

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The RAG system supports multi-language knowledge base and cross-language search, using Google text-embedding-004 (768d, padded to 1536) as primary and OpenAI text-embedding-3-small (1536d) as fallback.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The SimplifiedMemorySystem provides unified, context-aware conversation memory with a 3-minute TTL, agent isolation, and automatic cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Monitoring and alerting are implemented via real-time dashboards and webhook endpoints, tracking RAG performance, cache efficiency, API usage, and error rates.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: All admin and management interfaces (e.g., knowledge base, categories, metadata templates) are accessible via dedicated API endpoints and require proper authentication.

src/mastra/agents/enhanced-qa-agent.ts (3)

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The EnhancedQAAgent uses detectSpecificRequest() to identify specific user requests and limits responses to essential information, prioritizing precision over completeness.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: The SimplifiedMemorySystem provides unified, context-aware conversation memory with a 3-minute TTL, agent isolation, and automatic cleanup.

Learnt from: CR
PR: terisuke/engineer-cafe-navigator#0
File: CLAUDE.md:0-0
Timestamp: 2025-06-24T22:31:34.870Z
Learning: Memory-aware question handling: The system detects memory-related queries (e.g., 'remember', 'さっき', 'what did I ask') and uses conversation history instead of knowledge base search for responses.

🧬 Code Graph Analysis (7)

src/mastra/voice/google-cloud-voice-simple.ts (1)

src/utils/stt-corrections.ts (2)

applySttCorrections (82-128)

adjustConfidenceAfterCorrection (157-170)

scripts/test-stt-corrections.ts (1)

src/utils/stt-corrections.ts (3)

applySttCorrections (82-128)

likelyContainsMisrecognition (135-147)

CORRECTION_RULES (17-73)

src/test/test-memory-storage.ts (2)

src/lib/simplified-memory.ts (1)

SimplifiedMemorySystem (14-457)

src/lib/supabase.ts (1)

supabaseAdmin (78-82)

src/mastra/agents/realtime-agent.ts (1)

src/utils/stt-corrections.ts (1)

applySttCorrections (82-128)

src/lib/stt-correction-monitor.ts (1)

src/lib/supabase.ts (1)

supabaseAdmin (78-82)

src/lib/simplified-memory.ts (1)

src/lib/supabase.ts (1)

supabaseAdmin (78-82)

src/utils/stt-corrections.ts (1)

src/lib/stt-correction-monitor.ts (1)

SttCorrectionMonitor (19-196)

🪛 Biome (1.9.4)

src/app/api/qa/route.ts

[error] 62-62: Other switch clauses can erroneously access this declaration.
Wrap the declaration in a block to restrict its access to the switch clause.

The declaration is defined in this switch clause:

Unsafe fix: Wrap the declaration in a block.

(lint/correctness/noSwitchDeclarations)

src/lib/stt-correction-monitor.ts

[error] 19-196: Avoid classes that contain only static members.

Prefer using simple functions instead of classes with only static members.

(lint/complexity/noStaticOnlyClass)

[error] 47-47: Using this in a static context can be confusing.