Skip to main content
🔬 Research beta · specialists · non-diagnostic

Chat guidance for OTC choices

Rule-based scoring × LLM assist. Organizes OTC options and cautions from symptoms and context. This beta lets specialists evaluate live chat behavior—not a substitute for diagnosis or prescribing.

  • Chat-first intake — describe symptoms in plain language
  • Rule-based ranking — rationale, overlap warnings, and care guidance
  • Ten languages — Japanese, English, Chinese, Korean, Vietnamese, Nepali, Portuguese, Tagalog, Indonesian, Spanish (DeepL / Amazon Translate)

01

Three pressures on the ground

Aging populations, language barriers, and e-commerce are shifting demand—and strain—on pharmacies and shoppers alike. Gaps in counseling time can prolong uncertainty and unsafe choices.

Language barriers

Visitors and residents may struggle to consult in their language and choose appropriate OTC products.

Staffing pressure

Pharmacies have limited time for counseling, leaving some users uncertain.

Hard to choose

Users may not know which OTC fits their symptoms and worry about safety.

02

Three steps to use

①

Describe symptoms

Type symptoms and context in plain language—no medical jargon required.

②

See organized options

Drug DB, pharmaceutical rules, and rule-based scoring are core; the LLM assists NLU and questions.

③

Use as reference

Review cautions and when to seek care; consult a pharmacist or clinician before buying or taking medicine.

Live Sage Terrace UI (beta): symptom input → AI analysis → recommendations and comparison.

03

Four strengths

Safety by design

Rule-based core; refers to in-person care when symptoms are serious.

Hybrid recommendation

Rule-based scoring drives drug selection; the LLM assists NLU and questions only.

Ten languages

Japanese, English, Chinese (Simplified/Traditional), Korean, Vietnamese, Nepali, Brazilian Portuguese, Tagalog, Indonesian, and Spanish via DeepL / Amazon Translate.

Accessibility

Font sizing, text-to-speech, and collapsible sections; WCAG AA–oriented contrast.

04

Use safely

This tool does not replace medical care, diagnosis, or prescribing.

  • Seek in-person care promptly for severe or prolonged symptoms.
  • In Japan, call 119 for emergencies and use public helplines or clinics as needed.
  • If unsure about displayed information, consult a pharmacist or clinician before use.
  • Share pregnancy/breastfeeding, pediatrics, other medicines, and allergies when you can in chat.

Escalation to a pharmacist may be available depending on the deployed app version.

05

Technology stack

Beta building blocks. Rule-based logic is core for drug selection; the LLM assists.

AWS Enterprise Cloud

システム全体アーキテクチャ図(AWS)

ユーザー接点からECS Fargate、AI/LLM、マネージド分析、監視までを網羅した高可用性構成

ユーザー
LINE
Browser
ネットワーク・セキュリティ
AWS WAF
Application
Load Balancer
Amazon
CloudFront
Amazon S3
医薬品データベース
アプリケーション
ECS Express
ElastiCache
AI・LLM
Jev
OpenAI
Qwen
Kimi
AI/ML(AWS Managed)
OpenSearch
Serverless
Bedrock KB
Amazon
Personalize
Comprehend
Medical
Amazon
Polly
Amazon
Translate
運用・監視・セキュリティ
Amazon
CloudWatch
AWS
Budgets
Amazon
SNS
AWS
Lambda
Secrets
Manager
AWS IAM
外部サービス
Neon
PostgreSQL
Cloudflare
R2
GitHub
Actions
Text is not SVG - cannot display
AI & LLM

AI・LLM 基盤

  • jev
    Jev 判定
  • openai
    OpenAI GPT-5.4系(文章生成)
  • qwen
    Qwen 3.8 max / flash(高速トリアージ)
  • kimi
    Kimi K3(冗長フォールバック)
AWS Managed AI

AI/ML(AWS Managed)

  • Bedrock KB 添付文書ナレッジベース
  • OpenSearch Serverless ベクトル検索
  • Personalize 嗜好・適合レコメンド
  • Comprehend Medical 医療用語エンティティ抽出
  • Amazon Polly 自然音声合成 (TTS)
  • Amazon Translate 多言語翻訳 (10言語)
App & Network

アプリケーション・配信

  • AWS ECS Fargate FastAPI コンテナ実行
  • Amazon ElastiCache 高速セッション・キャッシュ
  • AWS WAF Web不正通信・ボット遮断
  • Application Load Balancer 冗長負荷分散 (ALB)
  • Amazon CloudFront グローバルエッジCDN配信
  • Amazon S3 静的アセット・データ保存
Ops & External

運用・監視・外部連携

  • CloudWatch & Budgets 構造化ログ・費用監視
  • AWS Lambda & SNS 自動アラート・イベント通知
  • Secrets Manager & IAM 暗号化認証・権限管理
  • pmda
    PMDA データベース 公的添付文書・相互作用
  • neon
    Neon PostgreSQL サーバーレスDB
  • cloudflare
    Cloudflare R2 & Tunnel ゼロトラスト安全トンネル
  • github
    GitHub Actions & Pipeline 安全自動CI/CD
  • line
    LINE Messaging API 公式LINEチャット対応
More detail ▾
  • Recommendation: proprietary rule-based scoring (symptoms, efficacy, age, interactions, etc.)
  • Multi-agent: with LLM_AGENT_ENABLED, ChatOrchestrator hands off after triage (Triage, Physical, Concierge, etc.)
  • Frontend: HTML/CSS/vanilla JS (responsive)
  • Logs: JSONL structured logs (access, performance, security)

06

Operations

Non-profit academic beta trial (per app overview documentation).

Source control

GitHub versioning; CI/CD via AWS CodePipeline.

Production & staging

AWS ECS Fargate (ap-northeast-1). Secure and high-availability delivery via Cloudflare Tunnel.

Database

Neon PostgreSQL (serverless) for sessions and global state.

Logs & monitoring

Amazon CloudWatch Logs + JSONL structured logs; access, performance, and security monitoring.

Limited to healthcare, government, research, and pharmacy specialists—not a general public launch.

Try the specialist beta

Explore rule-based recommendations and LLM-assisted chat in the live UI.