Case StudyAri Copilot

Revolutionizing UI Navigation with an MCP-Powered AI Voice Assistant

Services

  • UX-Driven AI

  • UI Design

  • React Development

  • OutSystems Development

Deliverables

  • Hight Fidelity Mockups

  • Interactive Prototype

  • React Web AI Browser

  • Extension

Client

  • Confidential

Learn how a new AI-powered extension is reducing by 70% recruiter stress by simplifying candidate search, turning hours of manual filtering into a single voice command.

Project Context

Finding Candidates Can Be a Voice Command

The client's recruiters were drowning in data. With a growing volume of candidates, job listings, companies, and requirements all living inside the platform, finding the right person for the right role meant navigating through multiple pages, stacking filters one by one, and repeating the process for every new search. The day-to-day was exhausting and time-consuming, not because the data wasn't there, but because getting to it required too many steps.

Mediaweb was brought in to explore how AI could change this. The goal wasn't to rebuild the platform, it was to sit on top of it and make it smarter. To give recruiters a faster, more natural way to interact with the data they already had.

Challenges

Filtering Were Slowing Recruiters Down

  • Filtering Were Slowing Recruiters Down

  • High cognitive load from juggling multiple screens and criteria simultaneously

  • Difficulty cross-referencing data across different sections of the platform

  • A growing database making manual search increasingly harder to manage over time

our approach

Exploring, Testing and Iterating

Was A Non-Linear Process

1. Start With Finding What's Possible


This project didn't use a linear process, it began with a question: what can we actually build? The team spent the first weeks in a technical exploration phase, investigating how the client's existing database was structured and what data could realistically be exposed to an AI model. Without understanding what the data looked like, any concept would have been built on assumptions.

2. Iterate, Iterate and Iterate

Unlike a traditional project where you define, design, then build, this one moved in cycles. The team would explore a technical possibility, sketch out an interface, test it against real data, discover a constraint, redesign, and test again. This happened repeatedly and intentionally. The iteration cycles kept the work grounded in what was real and buildable, while continuously validating value with the client along the way.

3. The Data-First Architecture

An early approach attempted to give the AI the ability to "see" the interface, identifying UI elements in real time to guide navigation. In practice, the cognitive load on the model was too high, causing noticeable lag and making the experience feel sluggish. The pivot was instead of teaching the AI to observe the UI, the team built a direct connection to the database using the Model Context Protocol (MCP). This removed the visual layer entirely for data tasks, making responses near-instant and 100% accurate to the system's live state.

4. Growing the Tool Together

Launching the first functional version of Ari Copilot wasn't the finish line, it was the starting point for a new kind of collaboration. Once the client's team began using the extension in their daily work, real feedback started coming in, small friction points, new use cases, edge cases that only surface when a tool meets real-world conditions. To help the client's internal team get up to speed quickly, a technical onboarding document was produced, covering how the extension works, how to interact with it, and what to expect as the tool evolves.

Solutions

What We Built to Transform Recruiter Workflows

1. Conversational Candidate Search

Before Ari Copilot, finding candidates for a role meant opening the jobs page, applying filters for company, job title, requirements, and location and repeating this every time. Now, a recruiter types or says "Find me candidates for a Senior Developer role in Lisbon" and the results appear immediately.


2. "Hey Ari"

To make the experience truly hands-free, the assistant can be activated by voice with the wake phrase "Hey Ari." Recruiters can initiate searches, check statuses, or navigate to specific records without stopping what they're doing. It keeps the focus on the work, not the platform.


3. AI-Generated Results

Results don't come back as plain text. The AI renders responses as interactive tables or candidate cards, formatted for quick scanning. Each card links directly to the relevant page inside the platform. Whether it's a list of candidates, a job summary, or a status update, the output is structured, visual, and immediately actionable.


4. Direct Database Integration via MCP

At the core is a custom MCP module built inside OutSystems. This connects the AI directly to the client's database, exposing business logic as callable tools. The AI can fetch candidate records, filter by criteria, check statuses, and return structured results, all without touching the visual UI.

The Result

Recruiters no longer start their day fighting through filters. They ask, and the platform answers.

Contact Us

Ready to rethink how your users interact with your product?

Every great product starts with a better experience. Let’s start yours.

Contact us