Back to selected work

Case study

EquitAI: human-centred AI for job applications

An open initiative I designed and built: an AI agent and automation prototype for CV building and job applications, starting from the barriers migrant and refugee women face in recruitment.

Visit the project site

Perspectives

Making AI, Working with AI, Experiencing AI

Method stages

AuditPathwayPrototype

Organisation and timeframe

Open initiative, started in the GRAILS 2025 programme at Elisava (Barcelona School of Design and Engineering) · 2025–present

The situation

Conventional recruitment systems under-represent skills and career paths that don't fit a standard template, including those of migrant and refugee women. At the same time, job searching is changing fast: applicants and employers are both turning to AI, and the process is becoming harder to understand for the people least able to absorb that complexity.

Audit

What I found

EquitAI grew from exploring the barriers faced by migrant women and people with non-standard career paths when they try to communicate their skills and experience.

Pathway

The pathway we designed

The project starts from a redesign question rather than a technology: what should AI automate, what should it only augment, and where must people stay in control? Three principles shaped the design: human judgement stays in the loop for decisions that matter; privacy, transparency and oversight are design requirements, not afterthoughts; and inclusion is a design criterion, including language, digital confidence and non-standard career histories.

Prototype

What we made tangible

A live, working tool that I designed and built: AI agents and automation that help people tailor CVs and cover letters and track their applications, while keeping the applicant's own voice and judgement at the centre. The aim is for AI to support human storytelling, not overwrite it.

I built the tool in Cursor with Next.js and TypeScript, using OpenAI and Anthropic models, and documented each design decision as an iterative Cursor prompt, so the reasoning behind every change is traceable. The agent finds relevant roles and drafts tailored CVs and cover letters, but nothing is sent without the applicant's approval. A separate review area shows each step the agent took and every item waiting for a decision, so its work can be checked, corrected or rejected. The version before the agent is kept so the product can be rolled back. Around the agent, the same principles apply throughout: no match percentages or scores that imply a certainty the system doesn't have, voice input only when the person chooses it, honest empty states, German-format CV export, and accessibility to WCAG 2.1 AA.

Open the EquitAI tool (external site)

Scale

What scaled, changed or stopped

EquitAI is extending the same question to workplace automation more broadly. A planned second case will look at redesigning an existing workplace workflow with AI: automation versus augmentation decisions, human-in-the-loop points and agents, guardrails and oversight, and accessibility checks.

What the organisation can do now

EquitAI shows that AI support for job applications can be built around the applicant's control instead of automation for its own sake. Its patterns, such as explanations instead of scores, approval before anything is sent, a visible record of what the agent did and opt-in voice, can be used by any organisation designing AI for people facing barriers, from careers services to public employment agencies.

My role

Concept, research, design and development of the AI agent and automation prototype. I designed and built the tool on my own, from positioning and research through product and interaction design, accessibility and implementation.