At Glassdoor, I led the development of a machine learning system to personalize job alert emails—transforming them from static listings to highly individualized recommendations. The goal wasn't just engagement—it was to reduce friction in the job search and help users discover better-fit roles faster. We started by building behavioral user profiles based on prior interactions: job views, saves, applications, and timing. Using matrix factorization, we captured latent preferences and encoded these into personalized vectors. This fed into an XGBoost model that scored user-job pairs based on job freshness, location fit, user engagement decay, and semantic similarity to prior clicks. To optimize timing, I layered in reinforcement learning to identify the best send time for each user based on historical open behavior. We deployed this system through Airflow to process millions of job-user combinations daily and generate dynamic emails tailored to each user's preferences and behavior. The impact: a 40% increase in apply-starts, a 25% reactivation of dormant users, and a notable drop in unsubscribes. But the deeper value was human: mid-career professionals, returning caregivers, and underrepresented job seekers were re-engaged with opportunities relevant to their goals. From a societal perspective, this system promotes economic mobility and inclusion. It bridges gaps in the labor market by surfacing meaningful opportunities to people who might otherwise be overlooked or overwhelmed. By reducing noise and highlighting fit, we helped restore dignity and momentum to the job search process—proving that machine learning, when human-centered, can deliver more than metrics: it can unlock progress.
One example of using machine learning to personalize an experience was when I worked on a campaign to optimize our e-commerce sales funnel. We used machine learning algorithms to analyze customer behavior, purchase history, and browsing patterns. The data allowed us to create personalized product recommendations for each user based on their unique preferences and past interactions. For instance, if a customer viewed certain products multiple times but didn't purchase, the algorithm would suggest similar or complementary items in future emails or site visits, nudging them closer to a sale. By tailoring the experience to individual users, we were able to significantly increase conversion rates and improve customer satisfaction. The key was constantly feeding the system with new data, so it could refine and adapt the recommendations in real-time, offering more relevant products with each interaction. The results were measurable — we saw a noticeable boost in customer engagement and sales from personalized recommendations.
In a recent project, I applied machine learning to customise an e-commerce shopping experience. Collating data per individual user, such as past purchases, browsing, and search queries, I set up a collaborative filtering system of recommendations. This system would look at the buying patterns of target consumers among other users with close preferences and then predict the items which most appeal to each target user. Furthermore, natural language processing helped create a robust search mechanism that could be utilised to find products more intuitively based on consumer queries. The model continuously updates recommendations in real time, ensuring relevance based on user interactions. This data-driven approach enhances personalisation, increasing user engagement and conversions as product suggestions align with individual preferences. Ultimately, this results in greater customer satisfaction and loyalty, highlighting the role of machine learning in crafting personalised experiences.
I once worked on a project where we used machine learning to personalize product recommendations on an e-commerce platform. By analyzing user behavior data—like past purchases, browsing history, and time spent on pages—we trained models to predict what products each customer would be most interested in. Instead of showing generic recommendations, the system tailored suggestions in real-time, increasing relevance for each user. One key example was for repeat customers: the algorithm identified patterns in their preferences and suggested complementary products they hadn't explored yet. After implementing this, we saw a 20% boost in average order value and a noticeable increase in customer engagement. Using data in this way allowed us to create a more personalized, satisfying shopping experience, proving how powerful machine learning can be in driving both user satisfaction and business growth.