Deep-dives on the systems I've built — ranking, entity resolution, agentic AI, and ML infrastructure.
How I architected a 5-layer system where LLMs plan and deterministic policy engines execute — zero hallucination-induced regulatory breach, by design.
From cohort-based defaults to true user-level personalization at Cars24 — +150% Click Recall@50, +16% buyer conversion, and solving cold-start for 35% of the user base.
Multi-stage canonicalization matching scraped entities to ~900M LinkedIn person nodes — ground-truth-first methodology with tiered candidate filtering under live traffic.