Swiftli Smart HR – Native HR software for macOS

Personnel records, shifts, recorded working time and payroll concern the same people but are handled from different perspectives. Smart HR is designed to connect these tasks in one working environment. It focuses on Swiss small and medium-sized businesses, particularly hospitality, with the authoritative business records stored on their own Macs.

We develop the interaction design and native application with Swift and SwiftUI. This includes organising the product around operational tasks, creating understandable working views and integrating local language models. Data storage and synchronisation within the local network are planned around how the application is used.

Roles and module access distinguish management tasks from employees' own views, for example. The interface needs to support both an overview of the business and work on an individual record.

The employee area brings personnel records and related documents together. The planner arranges shifts in a grid of employees and days, with templates and conflict notices supporting the work. Weekly and monthly plans can be exported as PDFs. Time recording, payroll views and reports have their own areas within the same Mac application.

• Personnel records and a document centre
• Shift grids, templates and planning-conflict notices
• PDF output for weekly and monthly schedules
• Dedicated views for time, payroll and business reports

The Smart Assistant translates written workplace rules into structured planning constraints. The August 2026 development account describes Qwen3-4B-Instruct-2507 with four custom LoRA adapters for German, French, Italian and Romansh. The model is quantised to four bits and runs locally through Apple MLX. The base model is pinned to a fixed revision; the base model and adapters are verified using SHA-256 checksums.

This work brings data preparation, adapter training and checked outputs into the application. A separate solver handles the actual planning. The account also identifies gaps in rule enforcement at that time, documenting a specific stage of development.

Local LLM integration includes work on quantisation and caching, with tests on Apple Silicon and consideration of devices such as the MacBook Air M1. Prefix caching reuses matching precomputed input states; result caching concerns completed outputs. Their suitability and validity are assessed separately for each workflow.

Time to first token, subsequent generation speed, memory use and output quality are considered together. Measurements with an empty cache must remain distinguishable from repeated inputs. This experience informs the planning of custom AI features.

Swiftli Smart Companion complements the family with mobile areas for Today, Schedule, Time and Inbox. The design responds to the device. Client projects receive their own requirements analysis and an appropriate visual language.

The product and module pages show the developed application through specific working views and functions. Our development account also explains the decisions behind the language adapters. Together, they show how we combine interface design, local architecture and AI in a larger software project.

The Smart HR product website currently presents the app for preorder and describes the publication status and approvals still outstanding. The schedule and chat views in our gallery are web visualisations of the native interface.