Software Documentation

This session presents a practice-driven approach to CLI-based documentation development. Using the command-line capabilities of OpenCode CLI, we have built a custom skill system. By combining standardized documentation templates with unified style guidelines, we have created a full-fledged documentation generation pipeline that supports intelligent restructuring and rewriting of technical content, producing structured and standards-compliant output. The goal is to explore a lightweight and reusable model for efficient technical documentation production, and to help enterprises build more standardized documentation development processes.

In 2025, with Huawei’s Ascend and Kunpeng platforms fully open-sourced—supporting users in deep customization and independent development, and jointly building an ecosystem—we have established an open-source-oriented information system and intelligent writing capabilities. Through documentation system design, writing mechanisms, and the definition and construction of open collaboration, while also introducing AI to enable intelligent document generation, quality inspection, and gating, we have significantly improved the quality and efficiency of open-source information development, meeting the needs of developers in co-building the ecosystem. Among these, the practical experience in AI-based intelligent document generation, quality inspection, and gating can serve as a reference for documentation efforts across various open-source communities.

Despite the rapid iteration of high-tech products driven by LLMs, the methods for creating, maintaining, and consuming the accompanying documentation have struggled to keep pace. Traditional methods may improve efficiency to some extent, but when dealing with complex products, documentation professionals still need to deeply understand product technical details and vast documentation structures to accurately locate and modify content, leading to significant bottlenecks in documentation maintenance.

To address this challenge, we have developed an AI-powered multi-agent solution for technical documentation, leveraging the powerful text generation capabilities of LLMs, semantic search technologies through embedded models, and agents with proactive thinking and execution abilities. This solution addresses pain points in document synchronization, review, proofreading, and operational maintenance, pushing these processes toward automation and intelligence. As a result, documentation transitions from a passive, static asset to an “active” resource capable of self-reviewing and continuously evolving.

As AI reshapes the landscape of technical writing, the issue of “factual hallucinations” in large language models (LLMs)—such as fabricated content, misattribution, and over-editing—has become a key barrier to their scalable application in high-accuracy scenarios.

This presentation proposes a new role paradigm for technical writers evolving into “Model Context Engineers.” Rather than simply writing text, this role involves systematically constructing, optimizing, and managing the trusted context that models rely on for generation.

This presentation provides an in-depth analysis of the underlying causes of hallucinations in technical documentation generation and defines the competency model and methodology for this new role. It introduces a three-tiered “Knowledge-Generation-Validation” anti-hallucination framework, detailing how to systematically build, optimize, and manage the input context for models, along with a feedback mechanism to create a closed-loop approach for mitigating hallucinations.

1. Establishing an AI-driven knowledge flywheel system: Connecting the full lifecycle of user insights, knowledge creation, knowledge accumulation, and scenario-based application to address broken links in traditional knowledge management.

2. Standardizing the knowledge flywheel process: Defining standardized workflows for knowledge demand management, classification, and grading to ensure systematic accumulation of knowledge assets.

3. Developing an AI-assisted tool platform: Leveraging AI-powered intelligent search to overcome timeliness limitations in knowledge management and enable automated circulation of knowledge assets.

4. Scenario-based closed-loop management: Building a repeatable knowledge management model through cross-department collaboration and scenario-driven operations.

Engineering teams struggle to balance documentation quality with development velocity. Manual documentation processes create bottlenecks, while outdated content leads to knowledge gaps and increased support burden. This presentation explores a practical approach: developing an AI agent that integrates directly into existing engineering workflows.

The speaker shares firsthand experience building and deploying an AI agent that operates within the software development lifecycle. The presentation covers the complete development journey from concept to production-ready system, demonstrating how an AI agent connects with version control, responds to code changes, and produces consistent output without disrupting established team processes.

Attendees learn the technical foundations of agent development including API integration with large language models, event-driven architecture design, and CI/CD pipeline automation using GitHub Actions. The talk addresses critical implementation challenges such as authentication handling, error recovery, output quality control, and managing edge cases.

The presentation emphasizes practical integration patterns that allow a documentation AI agent to complement rather than complicate engineering workflows. The speaker discusses how to identify appropriate automation opportunities, set realistic expectations, and measure agent effectiveness.

Participants gain actionable knowledge for evaluating, building, and deploying AI agents within their own engineering organizations.

As Generative AI evolves from passive chatbots to active agents, the role of the technical writer is shifting from content creator to information architect. This session introduces a cutting-edge, privacy-first workflow: the Automated Documentation Reviewer, built on the Model Context Protocol (MCP).

We will demonstrate a “Docs-as-Code” pipeline where an autonomous agent monitors a documentation repository, acting as a tireless preliminary editor. Unlike cloud-dependent tools that risk IP leakage, this workflow prioritizes Data Sovereignty. We will use various local models—such as DeepSeek R1 and other frontier labs’ models—to run locally, ensuring sensitive technical data never leaves the corporate firewall.

The architecture is designed for edge deployment with optional integration of larger models hosted on company servers through accessible API endpoints.

This session provides a guideline for enterprises seeking high-speed, compliant, and secure documentation engineering—eliminating reliance on third-party APIs while advancing technical writing standards.