Internal Chat System Development
European Consulting Firm | Internal Project
Overview
Designed, developed, and operated an internal inquiry response chat system utilizing generative AI and RAG (Retrieval-Augmented Generation) technology based on Azure cloud environment. Customized based on the Azure official reference implementation (azure-search-openai-demo) to achieve high-precision Q&A for internal knowledge base. Received the 2024 Corporate Award in recognition of this project's achievements.
Architecture
Built frontend with React/TypeScript, achieving UI compatible with both PC and mobile through responsive design. Adopted Python (Quart async framework) for backend, implementing real-time answer generation through streaming response. Built a multi-stage search pipeline combining high-quality natural language generation with Azure OpenAI Service (GPT-4) and vector search, hybrid search, and semantic ranker with Azure AI Search.
Built a structured analysis pipeline for PDF, Word, PowerPoint documents using Azure Document Intelligence as document processing infrastructure. Supported various document formats including internal regulations, manuals, and FAQs, automating from upload to Azure Blob Storage to automatic reflection in search index through Integrated Vectorization.
CI/CD & DevOps
Designed and built a full-fledged CI/CD pipeline using Azure DevOps. Achieved automation of build, test, and deployment with Azure Pipelines, establishing continuous delivery from development to staging to production environments. Introduced declarative management of Azure resources through Infrastructure as Code (Bicep/ARM Templates), ensuring reproducibility and consistency of environment construction.
Established source code management with Azure Repos and branch strategy (Git Flow), establishing a Pull Request-based code review process. Introduced task management and sprint planning with Azure Boards, spreading agile development practices throughout the team.
Operations & Monitoring
Implemented comprehensive application performance monitoring with Application Insights. Collected custom metrics including user satisfaction ratings (thumbs up/down), answer accuracy, and response time, building operational dashboards with Azure Log Analytics and KQL queries.
Gradually improved answer quality through prompt tuning and continuous optimization of search parameters based on user feedback. Implemented UX improvements including multi-turn conversation support, citation display, and thought process visualization, continuously improving user experience.
Achievement
Made significant contributions to improving the efficiency and quality of internal inquiry responses, receiving the 2024 Corporate Award in recognition of these achievements. This system has become established as a standard internal tool, with continuous use and improvement.