March 2026 Summaries

Rasepi's translation system is designed to produce translations that reflect the unique vocabulary and tone of a company, going beyond generic translation tools that often produce technically correct but tonally inaccurate results. It uses a paragraph-level tracking system, reducing translation costs by only retranslating changed sections, and maintaining consistency across languages. The platform allows for the creation of glossaries and style rules that ensure terminology is translated as intended and documents follow company-specific formatting and tone. This personalized approach supports both machine and human translations, where machine translations handle routine updates, and human translators focus on culturally nuanced content. Rasepi also supports unique content per language, allowing localized adaptations without being overwritten in future translations. This combination of glossary terms, style rules, and custom instructions ensures that translated documents sound like they were written by the local team, maintaining consistency and trust in multilingual documentation.

Rasepi's translation engine sets itself apart by focusing on translating individual blocks rather than entire documents, allowing for efficient use of resources and customization per tenant. When a document is edited, the system identifies changes through content hashing and only retranslates altered blocks, applying tenant-specific glossaries, style rules, and formality preferences. Glossaries, treated as mutable runtime artifacts in DeepL, are managed locally and synced when necessary, while style rules can be updated in place, offering flexibility for iterative refinements. The engine respects human edits by marking them as 'stale' rather than overwriting, and translations are orchestrated based on triggers that dictate timing, such as immediate retranslation or translation upon first access. This architecture not only reduces costs but also maintains consistency across translations by integrating terminology, style, and formality settings, ensuring that machine-generated translations align closely with the company's language standards. The same engine also supports Rasepi's conversational interface, ensuring consistent language quality in both written and spoken interactions.

Global companies often face challenges in maintaining consistent and up-to-date documentation across different languages and regions, primarily due to the inefficiencies of traditional documentation platforms which require separate copies for each language. This approach leads to multiple versions of the same document, each with shared global content and local-specific information, resulting in inconsistencies and outdated information. The article highlights the pitfalls of the "copy-and-localise" method, where local versions are often neglected, creating a maintenance burden and leading to discrepancies in shared content. A solution offered by Rasepi is to streamline documentation by maintaining a single document with shared content automatically translated into multiple languages, while local content remains unique to each language version. This ensures that all versions are current and reduces maintenance overhead, allowing local teams to manage their specific content without interference from global updates. This method not only enhances efficiency but also ensures that new hires receive accurate and relevant information tailored to their location, improving the overall onboarding process and documentation management in multinational organizations.

In multinational companies, the challenge of maintaining up-to-date and culturally relevant documentation across different languages is significant, often resulting in outdated or inadequate resources for non-English speaking teams. Traditional approaches to localisation, which involve translating entire documents, are costly, slow, and fail to accommodate unique regional content. Rasepi addresses these issues by implementing a block-level localisation system that translates individual sections, allowing for independent updates and the inclusion of unique, culturally specific content without overwriting changes. This method significantly reduces translation costs, enhances document relevancy in each market, and promotes trust in the documentation system, ultimately ensuring that every language version is a valuable and dependable resource for teams worldwide.

This continuation of a series on content freshness highlights why relying solely on expiry dates for documentation review is inadequate, emphasizing the necessity of continuous monitoring to maintain document accuracy and relevance. While expiry dates assign accountability by setting specific review dates, they fall short because documents can become outdated rapidly due to factors like broken links, changes in related content, readership drops, and outdated translations. The text introduces the concept of freshness scoring, which evaluates document health continuously using multiple signals, such as link health and readership trends, to catch issues as they arise. This approach complements expiry dates by providing immediate feedback and ensuring that documents remain trustworthy, especially as they are increasingly consumed by AI systems and other automated tools. The integration of both systems ensures that documents are reviewed regularly while also detecting changes or issues in real time, preventing the propagation of outdated or incorrect information.

Engineering teams often face challenges with outdated documentation, leading to the problem of "freshness" where documents may not reflect current practices or systems, especially as AI tools increasingly rely on these documents for information. The traditional approach to documentation, which assumed periodic updates by humans, is no longer sufficient, as AI assistants cannot discern outdated information and confidently provide incorrect answers. The concept of freshness extends beyond recent edits to include review status, link validity, readership, contextual changes, translation alignment, and community feedback, all contributing to a document's trust score. This issue is exacerbated by the growing reliance on AI, distributed global teams, and increased compliance demands, necessitating a freshness-first approach in documentation management. This involves setting mandatory review dates, automated health monitoring, and transparency in freshness scoring, with documents needing to continuously earn trust by maintaining accuracy and relevance. The cost of ignoring freshness can lead to confusion for new hires, misinformation from AI tools, compliance risks, and a general loss of trust in documentation systems. A practical starting point involves auditing key documents, setting review dates, and ensuring that AI tools source information from up-to-date documents, with visibility of freshness scores to encourage maintenance. The approach advocated by Rasepi integrates these principles into a platform that prioritizes documentation freshness as a core element of organizational knowledge management, paving the way for reliable AI-assisted solutions.

Deploying AI assistants on internal knowledge bases often leads to trust issues due to their inability to discern between current and outdated documents, primarily because AI systems, including large language models and retrieval-augmented generation systems, rely heavily on semantic relevance rather than reliability. This results in confidently incorrect answers when outdated or deprecated information is presented as authoritative, which can mislead users, particularly new hires, and erode trust in AI tools. The solution lies not in refining AI models but in enhancing the metadata of the documents themselves, such as implementing freshness scores, expiry statuses, classification labels, and language-level signals to ensure that AI systems can better differentiate between trustworthy and stale content. By providing documents with this rich, accurate metadata, tools like Rasepi enable AI systems to deliver more reliable answers by filtering out outdated or irrelevant information, thereby addressing the underlying data quality problem rather than attempting to modify AI behavior.

Human communication has traditionally relied on speech, an inherent skill developed long before written language, making conversation often feel more natural and less cognitively demanding than reading, which requires active decoding and interpretation. This distinction becomes particularly relevant as more information is stored in complex documents, which can overwhelm individuals, especially under stress or time constraints. In contrast, conversational interfaces provide a sequential, interactive experience that mirrors natural human dialogue, offering immediate feedback and reducing the mental load by presenting relevant information first. While reading is essential for depth and precision, especially when accuracy is crucial, conversational access to information is becoming increasingly popular as it aligns with the brain's natural preference for processing uncertainty through dialogue. The future of information systems lies in a balanced integration of both reading for comprehensive understanding and conversation for quick, intuitive access, ensuring reliability and depth are maintained without sacrificing the human-centered approach of conversational interaction.

Confluence and Notion are successful documentation platforms that have historically addressed the needs of their users by providing centralized spaces for writing, organizing, and sharing knowledge. However, as the world moves towards AI-driven systems, these platforms face challenges because they were originally designed for a pre-AI era, where the primary goal was to make documentation searchable and navigable for human users. The rise of AI demands that documentation be machine-readable, freshness-aware, and structured for automation, introducing new requirements that older platforms may struggle to retrofit. While Confluence and Notion are integrating AI features such as improved search and writing assistance, the core architecture of these platforms remains rooted in the past, making it difficult to fully embrace the AI-native approach that new platforms can adopt from inception. These newer platforms have the advantage of designing their systems with current technological and user expectations in mind, allowing them to more effectively manage knowledge as a trusted system and accommodate the evolving demands of AI. Despite the challenges, Confluence and Notion possess the resources and capabilities to adapt over time, but they must navigate the complexities of transforming their established systems to meet the requirements of the AI era.

Rasepi's platform aims for genuine extensibility through a well-structured architecture comprising plugins, action guards, and pipelines. Plugins are implemented using a modular design, allowing each to declare services, routes, and UI contributions, while being registered and scoped efficiently. Action guards serve as a control layer, determining whether actions, including core capabilities, can proceed by intercepting operations and evaluating conditions. The system's action pipeline executes operations in a deterministic manner, ensuring that all necessary guards are evaluated before actions are allowed, thus maintaining security and reliability. Additionally, the platform incorporates a block-level translation engine that optimizes costs by translating only modified document sections, while tenant isolation ensures strict data separation. This architecture supports adaptability to different workflows without compromising the platform's stability or security, thereby preventing potential disruptions in documentation processes.