AllyO - AI recruiting platform

Project Overview

The hard part of an AI product is rarely the model. AllyO's models were owned by a separate data-science team and exposed as APIs; my job was the layer where their output met a human being who had to make a hiring decision from it. Presenting a probability as a number is easy and dangerous - a recruiter who reads 0.78 as a fact makes a worse decision than one who sees no score at all. So the interface was designed around confidence: what to show, what to rank, and what to refuse to state plainly.

Challenges

• Surfacing automated candidate evaluations produced by models owned by a different team, against APIs I did not control.
• Presenting probabilistic model output so that recruiters could act on it without mistaking confidence for certainty.
• Making hiring-funnel trends legible across a large volume of candidates and stages.
• Keeping a large React/Redux single-page application maintainable as the product grew.

Solutions

• Led the front end of the React/Redux SPA surfacing automated candidate evaluations.
• Designed the interface layer for probabilistic model output - candidate scores, ranked results and hiring-funnel trends - in D3.js and Highcharts.
• Built visualisations that let recruiters act on model output without misreading confidence.
• Worked against APIs owned by a separate data-science team, coordinating the contract between the two.

Key Features

• Automated Candidate Evaluation - Model output surfaced in a form recruiters can act on.
• Confidence-Aware Presentation - Scores and rankings shown so that uncertainty stays visible.
• Hiring-Funnel Analytics - Trends across stages and candidate volume, in D3.js and Highcharts.
• Large-Scale SPA - React and Redux architecture sustained as the product grew.