Bin Duan

Head of Solutions for Huawei Cloud in Latin America, based in São Paulo. I build and share practical engineering assets around the 1 + 3 framework: Foundation + Application Modernization + Big Data + AI.

If you’re new here, start with About or browse the latest posts below.

Featured

• Deploying Qwen3.6-27B-W8A8 on Huawei Ascend 910B with vLLM Ascend

  Updated: 31 May, 2026

  A field-tested deployment guide for Qwen3.6-27B-W8A8 on Huawei Ascend 910B/A2 with vLLM Ascend v0.19.1rc1 and vLLM v0.19.1, covering CANN 8.5.2, Python 3.11, torch-npu 2.9, Triton Ascend, native kernels, ATB, graph mode, and the failure fixes needed to make deployment repeatable.

• Building a Multilingual AI Collections Agent for X Bank with GLM and LangGraph

  6 Apr, 2026

  How to build a multilingual AI collections agent with deterministic workflow orchestration, GLM-assisted classification and generation, retrieval-backed script selection, and session-aware negotiation logic.

• Building a Local Karmada Failover Demo: Multi-Cluster Kubernetes Traffic Switching on One Host

  5 Apr, 2026

  A step-by-step guide to building a single-host Karmada failover demo with two kind member clusters, a stable HAProxy entrypoint, and visible traffic cutover between Kubernetes clusters.

• Running LangGraph with openGauss: Integration Guide

  4 Apr, 2026

  Complete guide to deploying LangGraph with openGauss persistence layer - installation, configuration, and production-ready checkpoint management for AI agent applications.

• How We Successfully Started Qwen3-Coder-Next on Huawei Ascend 910B with vLLM-Ascend 0.17

  28 Mar, 2026

  Complete deployment guide for Qwen3-Coder-Next on Huawei Ascend 910B with vLLM-Ascend 0.17. Documents the working configuration after multiple failed attempts, covering CANN 8.5.1, W8A8 quantization, worker startup modes, and KV cache management.

• Mixed AI Compute Pools: Architecture Patterns for Heterogeneous GPU Infrastructure

  18 Mar, 2026

  A comprehensive guide to building production-ready mixed GPU pools. Explores capability-based segmentation, elastic burst patterns, multi-cloud federation, and real-world case studies including Huawei Ascend + NVIDIA hybrid deployments.

• Deploying GLM-5 W4A8 on Huawei Cloud Ascend 910B x8

  11 Mar, 2026

  A field report on deploying GLM-5 W4A8 on ModelArts with 8 Ascend 910B cards. We solved Python stack issues and got the API server ready, but hit a decode blocker at the Ascend runtime layer. This is a practical summary of what worked, what broke, and where the real problem turned out to be.

• From Zero to a Working Coding Assistant: Deploying Qwen3.5-9B on a Huawei Cloud 910B Notebook

  10 Mar, 2026

  A practical guide to deploying Qwen3.5-9B on Huawei Cloud ModelArts with Ascend 910B, including model selection, runtime setup, and opencode integration for coding assistance.

• Chilean Spanish ASR Model Benchmark Report: Comprehensive Evaluation of Open-Source Speech Recognition Models

  7 Mar, 2026

  In-depth benchmark of 7 state-of-the-art ASR models on Chilean Spanish. Voxtral-Mini-4B achieves 2.63% WER, Whisper delivers 9.5x real-time speed, and Fun-ASR offers best cost-efficiency. Complete performance analysis, deployment recommendations, and roadmap.

• From Rule Explosion to Runtime Discovery: Building a Universal Cloud Agent in One Day

  25 Feb, 2026

  How we eliminated 70,000 lines of hardcoded logic by letting the SDK tell us what it can do. A technical deep-dive into runtime discovery, LLM-first planning, and the 2-hour plan model rule that saves 20 hours of debugging.

• Building hc-agent: Lessons from Rapid Prototyping to Production

  25 Feb, 2026

  A technical post-mortem on building an AI-first cloud automation framework: from a rapid prototype built with Codex on a plane to a production-ready system, and the hard lessons learned along the way.

• Huawei OmniInfer PD Disaggregation on Ascend: Qwen3-VL Single-Node Deployment (1P1D)

  4 Feb, 2026

  Practical guide to deploying Qwen3-VL with Prefill-Decode separation on Ascend NPUs using OmniInfer. Covers minimal 1P1D setup, KV cache transfer, troubleshooting, and complete working scripts.

• Building an AI-Powered Telecom Marketing System: From Network Data to Personalized Campaigns

  4 Feb, 2026

  End-to-end guide: leverage Ookla Open Data + Huawei Cloud DLI + MaaS to build an intelligent marketing platform for telecom operators. Includes complete demo app with Streamlit + PyDeck visualization and AI-generated marketing scripts.

• Building a Multimodal Agent on Huawei Cloud Ascend: Qwen3-VL + LangChain

  3 Feb, 2026

  End-to-end guide to deploying Qwen3-VL multimodal inference server on Huawei Cloud Ascend 910B with OpenAI-compatible API, and building practical multimodal agents using LangChain with caching strategies.

• Deploying a Qwen3-VL OpenAI-Compatible API on Huawei Cloud Ascend (910B) with Docker + FastAPI

  3 Feb, 2026

  A comprehensive guide to deploying Qwen3-VL-8B vision-language model on Ascend 910B NPUs with an OpenAI-compatible FastAPI server, including container setup, load testing, and production deployment patterns.