Nền tảng luyện nói tiếng Anh tích hợp AI theo hướng sản phẩm thực tế: hội thoại theo chủ đề, phản hồi ngữ pháp, chấm phát âm chi tiết, theo dõi tiến bộ và vòng lặp ôn tập bằng flashcard.
- Giúp người học luyện nói đều đặn trong môi trường mô phỏng thực tế.
- Trả phản hồi nhanh, rõ ràng, có thể áp dụng ngay.
- Theo dõi tiến bộ theo phiên và theo thời gian.
- Tập trung vào trải nghiệm liền mạch: từ nói -> nhận feedback -> ôn tập.
- Chat theo chủ đề (IELTS, daily, business, ...).
- Gửi text hoặc ghi âm trực tiếp.
- STT tự động khi người dùng gửi audio.
- AI phản hồi bằng text và giọng nói (TTS).
- Highlight lỗi theo từng đoạn chữ.
- Trả câu sửa gợi ý, phân loại lỗi, giải thích ngắn.
- Lưu theo message để xem lại khi mở lại lịch sử.
- Tích hợp Azure Pronunciation Assessment.
- Điểm thành phần: overall, accuracy, fluency, completeness, prosody.
- Drill-down tới word/syllable/phoneme.
- Tổng số phiên, thời lượng luyện, streak.
- Biểu đồ xu hướng điểm và quy đổi dạng band.
- Gợi ý hướng cải thiện theo điểm yếu.
- CRUD deck/card.
- Upload media (image/audio) cho từng mặt thẻ.
- Ôn tập theo thuật toán SM-2 (spaced repetition).
- JWT auth, password policy mạnh, OAuth.
- Input/output guardrails (rate limit, injection filter, redaction).
- Audit logging và trace observability.
- React + Vite + TypeScript + Tailwind
- Các màn hình chính:
/chat: Voice Agent/dashboard: tiến bộ học tập/flashcards/*: quản lý và học flashcard
- FastAPI, REST API dưới
/api - Pipeline AI (LangGraph): preflight -> respond/tool -> TTS
- Service tích hợp:
- Groq (STT + LLM)
- ElevenLabs (TTS)
- Azure Speech (assessment)
- PostgreSQL: auth, conversation, grammar, pronunciation, flashcard
- MinIO: audio messages, flashcard media
- Redis: rate limiting
- Prometheus + Grafana (metrics)
- Elasticsearch + Kibana + Vector (logs)
Stack local gồm:
voice_agent_backendvoice_agent_frontendvoice_agent_postgresvoice_agent_redisvoice_agent_miniovoice_agent_pgadminvoice_agent_prometheusvoice_agent_grafanavoice_agent_elasticsearchvoice_agent_kibanavoice_agent_vector
- Docker Desktop
- Python 3.10+
- Node.js 18+
- Git
- (Khuyến nghị)
uvđể quản lý môi trường Python
git clone <your-repo-url>
cd A20-App-014-meCopy-Item .env.example .envBiến quan trọng cần cấu hình:
JWT_SECRET_KEYPOSTGRES_DB,POSTGRES_USER,POSTGRES_PASSWORDMINIO_ACCESS_KEY,MINIO_SECRET_KEYGROQ_API_KEYELEVENLABS_API_KEYVITE_API_BASE_URL
Biến tùy chọn cho pronunciation:
AZURE_SPEECH_KEYAZURE_SERVICE_REGION(hoặcAZURE_SPEECH_REGION)
docker compose up -d --buildKiểm tra dịch vụ:
docker compose pscurl http://localhost:8000/healthKết quả mong đợi:
{"status":"ok"}cd frontend
npm install
npm run devURL: http://localhost:5173
POST /api/auth/registerPOST /api/auth/loginGET /api/auth/mePOST /api/chat/respondPOST /api/chat/transcribePOST /api/assessGET /api/conversationsGET /api/conversations/{id}/messages-with-scoresGET /api/topics/get_categories_topicsGET|POST|PATCH|DELETE /api/flashcards/*
Ví dụ chạy nhanh một số nhóm test backend:
python -m pytest tests/test_security/test_security.py tests/test_api/test_routes.py -q
python -m pytest tests/test_api/test_schemas.py tests/test_ai_services/test_ai_services.py -q- Nếu thay đổi schema, reset volume DB trước khi seed lại.
- Với môi trường production/staging, bắt buộc dùng secret mạnh và endpoint storage/public URL đúng domain.
- Theo dõi metrics và logs trước khi mở traffic thật.
- Luồng hoạt động end-to-end của AI LinguAI
- Cụm Kubernetes trên Google Cloud
- Cơ sở dữ liệu production trên Google Cloud
- Quản trị Kubernetes bằng Rancher
- Pipeline CI/CD tổng thể
- Pipeline CI/CD trên GitLab
- Harbor container registry
- Logging pipeline (thu thập và đẩy log)
- Monitoring pipeline (metrics và giám sát)
- Log monitoring trên Kibana
- Trace/LLM monitoring trên LangSmith
- MinIO Object Storage
- Xem video demo: docs/demo/demo.mp4
- Nếu GitHub không phát trực tiếp trong tab preview, hãy click vào link trên để mở hoặc tải file video.











