At AST Consulting, I own match-making and recommendation systems across B2C workflows. I built two-tower retrieval and a LightGBM LambdaMART re-ranker, reducing manual matching effort by 65%.
I designed multi-objective ranking and a real-time feedback loop that updates user preferences from views, dwell time, likes and skips. Streaming feedback through Kafka into Redis cut serving latency by 40%.
For sparse-data users, I combined LLM-generated profile embeddings and content-based priors with collaborative-filtering signals. I also developed graph-based match scoring to reflect mutual interest.
I deploy ranking services and build their evaluation and observability systems, using offline metrics and online A/B tests to check that model improvements translate to product outcomes. In my projects, I’ve also built reciprocal matchmaking, e-commerce ranking and semantic retrieval systems.
