Integrating Conversational AI Into A Movie Streaming Platform: Architecture, Rag Workflow, And Empirical Evaluation
Từ khóa:
AI chatbox, movie streaming, RAG, information retrieval, OpenAI, AWS, Web Engineering, conversion analyticsTóm tắt
We describe the integration of an AI chatbox into CartoonToo, a web-based movie streaming platform backed by AWS (S3/CloudFront/Cognito) and Spring Boot services. The assistant is implemented with a retrieval-augmented generation (RAG) pipeline that grounds answers in the platform’s own knowledge sources—movie catalog metadata, payment and VIP FAQs, and active promotion policies. A compact intent router steers queries to task-specific prompts, while guardrails enforce domain scope and privacy. From a software-engineering viewpoint, we present an end-to-end Web Engineering workflow covering requirements, implementation, CI/CD, observability, and latency/cost controls (Caching, evidence prefetch). We conduct an empirical evaluation combining functional testing, personalization quality analysis, and performance benchmarking, measuring technical quality (human-judged answer accuracy, p95 latency), intent-recognition F1, and the effect of fast-path and fallback mechanisms on cost and reliability. The results show that the assistant reliably handles normal, abnormal, and boundary cases, keeps p95 latency under 2 seconds, and reduces LLM calls and operating cost via no-train ML and caching. We conclude with trade-offs among quality, safety, and cost, current limitations (coverage, Vietnamese/English handling), and planned extensions such as tier-aware personalization and smaller on-prem models.
