ENGINEERING CASE STUDY · FULL-STACK APPLICATIONS
mAble AI Automation Platform
Engineering case study.
LARAVEL 12 · REACT · TYPESCRIPT · INERTIA · SANCTUM · OPENAI · ELECTRON · JEST · GITHUB ACTIONS
A multi-tenant platform where teams generate and run custom automation tools, with secure company and user management.
- Role
- Senior full stack engineer
- Type
- Multi-tenant AI product
- Impact
- 70%+ less manual workload
- Source
- Private (employer-owned)
01 / OVERVIEW
The project
mAble is an AI-powered automation product for technical professionals. A Laravel 12 and React/TypeScript admin application manages companies, users, roles, invitations, saved tools and prompts, while an Electron and React desktop client turns plain-language requests into runnable automation tools. The product was built for an employer, so the source code is private.
02 / ARCHITECTURE
How it fits together
How the admin application and the desktop client work together.
CLIENT
Electron + React desktop client
Users describe a task in plain language. The client generates a layout and task logic, then renders the tool from a reusable component library.
Sends to: LLM provider, Authenticated API
A simplified view of the employer-owned product, based on its documented structure.
03 / CONTRIBUTION
My contribution
- Built backend and API functionality for company and user profile features, including authenticated endpoints for users, companies, tools, prompts and saved apps.
- Integrated LLM providers into production workflows, including request validation, decision routing, fallback handling and structured responses.
- Built authenticated REST, webhook and structured-payload integrations with validation and error handling.
- Reviewed, debugged and tested AI-generated JavaScript and TypeScript, and built evaluation workflows for accuracy, consistency and edge cases.
04 / ENGINEERING
Engineering notes
- Tenant isolation: middleware restricts users to their own company's data, with a dedicated test suite for tool and company isolation.
- Access control: Laravel Fortify and Sanctum handle authentication, with two-factor authentication, role and permission checks, active-user checks, and an activity log for auditing.
- Invitations: team members join through emailed invitations that are accepted on a dedicated page, rather than open registration alone.
- API boundary: Inertia pages and JSON endpoints keep the React interface separate from backend business logic.
- Desktop client: a Node main process calls the OpenAI API with a constrained system prompt to generate layout and task logic, then renders the result from a reusable component library (buttons, file pickers, sliders, progress bars, output console).
- Delivery: PHPUnit and Jest tests, linting, browser tests and deployment workflows run through GitHub Actions.
05 / OUTCOMES
Outcomes and context
- Automation and AI-assisted workflows reduced manual workload by 70%+.
TECHNOLOGY
Tools selected for the work.
- Laravel 12 with Sanctum and Fortify. Mature, batteries-included authentication, two-factor support, validation and queues, which suit a multi-tenant product where access control matters most.
- React, TypeScript and Inertia. A typed interface that stays close to the backend without a separate API layer for every page.
- OpenAI API. Turns plain-language requests into runnable tools, with a constrained system prompt to keep output predictable.
- Electron. The tools run on the user's own machine and need file access, which a browser cannot offer.
- PHPUnit, Jest and GitHub Actions. Tenant isolation and AI output both need regression protection, so tests run automatically on every change.