Jingyao Chen

About

Cofounder at MeowyAI who shipped a production multimodal (vision/voice/text) AI task manager using Gemini, tackling real-world issues like hallucinations, tool-calling safety, and RAG-based preference memory. Also built a production multi-agent RAG system orchestrated with LangGraph (and contributes to LangChain), with strong emphasis on latency optimization, observability (OpenTelemetry), and rigorous testing/evaluation including A/B tests and adversarial prompting.

Key Strengths

• Built and deployed multimodal AI task management system to production (Gemini vision/voice/text)
• Improved LLM reliability via structured output validation (Pydantic) and retry/prompt refinement to reduce hallucinations
• Designed robust tool-calling with idempotency keys and transaction logs for safe third-party API operations
• Deep RAG expertise: hybrid retrieval (BM25 + vector), reranking, filtering, and preference memory
• Production multi-agent orchestration with LangGraph (router + specialized subgraphs) and context management
• Performance/latency optimization in deep agent graphs using smaller models and streaming
• Strong evaluation and observability practice: node-level unit/mocks, query-doc benchmarks, A/B tests, OpenTelemetry tracing (latency/token/success)
• Reliability engineering mindset: failure modes upfront with Istio circuit breakers, timeouts, and traffic control
• User-trust driven product design: human-in-the-loop approvals, confidence scoring, and explainability based on stakeholder/user feedback
• Open-source contributor to LangChain

Experience

Education

Skills

Languages