I build Dify plugins, RAG integrations, and model-serving infrastructure, with upstream contributions to Dify, Xinference, and RAGFlow.

• RAGFlow: Connect Dify to RAGFlow for datasets, documents, chunks, retrieval, and memory workflows.
• Lingua: Detect languages with ISO language-code output and optional language scoping. Dify Marketplace
• MinIO: Read, upload, list, and inspect objects in MinIO-backed AI workflows.
• Knowledge: Use existing Dify knowledge-base operations inside plugin workflows.
• Moderation: Moderate Dify app input and output with configurable keywords, preset responses, and keyword masking.
• Juhe: Access weather, exchange rates, oil and gold prices, and stock-market data through Juhe APIs.

• skills: Reusable agent skills for the workflows I use and maintain.

Representative merged PRs, from retrieval performance to distributed model serving.

• Knowledge retrieval performance #42673: Cache parsed provider declarations, check query embedding caches before constructing models, and reuse rerank model metadata; the PR's 100-request benchmark reported 42.2% lower mean latency and 73.7% lower P95 latency.
• Chatflow startup performance #36773: Reuse request-scoped provider metadata and runtime objects to reduce repeated model resolution; the PR's local benchmark reported 41.3% lower mean workflow startup latency.
• Excel image ingestion #37104: Extract embedded images from .xlsx knowledge imports, store them as uploaded files, and preserve image-only rows for downstream indexing.

All merged Dify PRs

• Dynamic replica scaling #5426 / #5491: Add configurable scale-up and placement across multiple workers, with CPU/GPU allocation, failure rollback, and replica-state reconciliation.
• Supervisor restart recovery #4731 / #5250: Reconnect workers and rebuild replica state after supervisor restarts, including recovery from stale supervisor references.
• OpenTelemetry integration #4666: Introduce optional OpenTelemetry instrumentation with a configurable OTLP endpoint for model-serving observability.

All merged Xinference PRs

• Dify Official Plugins: Provider and model updates for SiliconFlow, DeepSeek, Volcengine, Azure OpenAI, and tool-call message handling.
• Dify Plugin SDKs and Plugin Daemon: Runtime improvements around app context, persistence, and storage accounting.
• RAGFlow: Model-provider additions and retrieval-system integration work.