shijiezheng

Project Description: Mindee is an AI mental wellness companion app built around intent recognition, long-term memory, and emotional support strategies. It provides personalized AI conversations based on CBT, Solution-Focused Brief Therapy (SFBT), and crisis intervention, with structured features such as MBTI profiling, Johari Window self-awareness, mood check-ins, growth tracking, reflection notes, and daily companion reports. Responsibilities: 1.Independently owned AI and backend development, working with psychological consultants to translate counseling methodologies into orchestrated dialogue flows and structured output standards 2.Built Agent invocation and async task dispatching with FastAPI microservices, Kafka, and Alibaba Cloud Model Studio (Bailian), decoupling real-time dialogue from persistence and batch processing 3.Designed multi-service backend architecture with independent PostgreSQL databases and JWT-based cross-service authentication to support horizontal scalability 4.Owned Agent context management, Skills architecture, MCP design, prompt design, RAG knowledge base construction, and continuous optimization 5.Built a long-term memory and user profiling system using Bailian Memory API, multi-window sessions, and local message storage; used MBTI, preference tags, and key conversation signals for personalized Agent routing 6.Designed a dual-Agent CBT workflow with detection and dialogue Agents, using emotion, stage, and intent signals to drive strategy-based prompt injection Abstracted CBT, SFBT, Johari Window, crisis intervention, mood check-ins, reflection notes, and growth tracking into orchestrated Agent workflows with structured results persisted to the database 7.Developed multi-user, multi-turn load testing scripts and established P95 / P99 latency and success-rate baselines; achieved 100% success rate in a 90-turn test scenario and covered core paths with pytest

Focused on applying LLMs to audio and content platforms, improving intelligent recommendation systems and chatbot interactions. Developed a context-aware dialogue assistant tailored for audio platform users, optimizing personalization and content delivery. Participated in prompt engineering and fine-tuning tasks for internal knowledge-based AI tools. Collaborated with product teams to align AI solutions with business needs, supporting product enhancement through data-driven iteration. Explored integration of memory modules and persona frameworks to enhance long-form conversation quality.

Participated in the development and deployment of AI applications using large language models (LLMs), focusing on prompt engineering, retrieval-augmented generation (RAG), and multi-turn dialogue optimization. Designed and built intelligent chatbots and internal tools to enhance business productivity, including an LLM-powered memory system for AI NPCs. Published an SCI paper titled MemoryRepository for AI NPC, presenting a modular memory architecture for long-term interactions in conversational agents. Contributed to model fine-tuning and custom dataset preparation, improving chatbot coherence, retrieval accuracy, and human-like interaction quality. Worked cross-functionally with product and engineering teams to integrate LLM solutions into real business workflows.