I&T Solution

1. Lowering the capability barrier. The solution improves public services by embedding AI capability directly into the Office interface that frontline staff use every day. Public-service roles commonly face the dilemma of systems being built but not actually used. By making natural language the sole interaction mode, the solution enables users to complete high-frequency tasks—drafting and format-standardising official documents, consolidating statistical reports and checking their definitions, producing briefing and presentation materials, and classifying and replying to citizen letters and service-ticket emails—without having to master formulas, macros, templates or layout rules. For new hires and staff changing roles, the ramp-up period from "zero experience" to independently producing compliant output is shortened from weeks to days, which meaningfully advances digital inclusion.2. Raising processing efficiency and output quality. The solution automatically routes more than one hundred native Office operations according to the host application (Excel, Word, PowerPoint and Outlook), and connects to external business systems through the MCP protocol, so that the entire chain—data retrieval, analysis, drafting, briefing and dispatch—closes within a single session, compressing the turnaround time for a single document from hours to the order of tens of minutes. At the same time, session persistence and context compaction keep long tasks stable; a three-layer pipeline of text extraction, page rendering and visual understanding handles scanned and photocopied documents; and screenshot-cropping self-checks verify chart overflow—together ensuring that AI output is of a quality that can pass acceptance. 3. Making governance and oversight quantifiable. The multi-tenant architecture allows a supervising authority to configure model whitelists, permission policies and usage quotas centrally, while subordinate units operate autonomously within that unified framework. Tenant-level usage and audit reports provide a data basis for oversight of public-resource consumption, cost allocation and performance evaluation. 4. Cost structure: clear, controllable and progressively decreasing. 4.1 One-off cost. This consists mainly of platform development and four-host adaptation. Because the four hosts share a single backend deployment and a single front-end build, with capabilities delivered through configuration rather than per-host development, the duplicated investment of building four separate systems is avoided by design. 4.2 Deployment and operating cost. Deployment cost is low—a single containerised package (Node service, Postgres, object storage and an Nginx gateway) is sufficient to serve all users across the four hosts—and configuration changes support hot reload, requiring no downtime or restart. Operating cost is markedly lower than the traditional pattern of procuring multiple separate tools, thanks to complete audit trails, centralised configuration management and health-check mechanisms. Training cost is sharply reduced by natural-language interaction, with dedicated software-operation training largely no longer required. 4.3 Model inference cost. The core variable during operation is model inference cost. The platform meters token consumption precisely across three dimensions—tenant, user and model—through a usage table, and inference can either be procured centrally by the platform or supplied by the customer through BYOK with their own keys or on-premise local models. This turns inference cost into a resource the customer already owns, approaching zero in local-model scenarios, and eliminates from the outset the price-increase risk of lock-in to a single model vendor. 5. Benefits to public services. 5.1 Fiscal benefit. Duplicated construction and duplicated procurement for equivalent capability fall significantly, and as the number of deploying units grows, the average construction and operating cost per user continues to be amortised, producing a scale effect of "build once, reuse across levels".

5.2 Workforce benefit. High-frequency, repetitive and format-heavy labour is taken over by the agent, allowing frontline staff to return their energy to analysis, decision-making and service delivery itself. At a few hours saved per user per month, an organisation of a hundred people can release several thousand person-hours a year, equivalent to several full-time employees, and this effect amplifies as usage frequency rises.

5.3 Demonstration benefit. The solution already offers containerised deployment packages, an enterprise installer, a multilingual interface and multi-currency billing, so it can be deployed centrally on a provincial platform or rolled out lightly at district and county level, providing the public sector with a replicable, transferable and sustainably evolving implementation paradigm for applying AI.