Selected work.
A look at the kinds of prediction and recommendation problems we take on, and how we structure them.
Cross-Platform User Interest Evolution Prediction
A hybrid model that fuses graph relationships, behavioural sequences and tabular business features to predict how user interests shift over time, and where they'll go next.
Discuss a similar projectProblem shapes we handle well.
Cross-platform ranking
Unifying signals from multiple products into one ranked, Top-K recommendation stream.
Cold-start handling
Sensible recommendations for new users and items before behaviour data accumulates.
Churn prediction
Early-warning models that flag disengagement in time to act on it.
Interest graphs
Modelling how users, items and interests connect, and how those links evolve.
Benchmark studies
Head-to-head evaluation of candidate models against strong baselines.
Model productionisation
Taking a validated research model through to a service your product can call.
Have a project in mind?
Bring the data problem, we'll help scope it into something measurable.