From LLM-powered QA automation to fraud detection at scale — each project is a proof point, not a proof of concept.
LLM-powered evaluation system replacing manual vendor review of customer service conversations — covering 32 products, 76 workflows, and multiple languages simultaneously using Gemini, RAG, and async batch inference.
Centralized platform for managing ML signals across support systems — six integrated layers covering ingestion, storage, quality monitoring, discoverability, and alerting for data scientists across teams.
Hybrid ML + heuristics personalization engine delivering context-aware help recommendations across support surfaces — with cross-product signal federation and privacy-compliant architecture.
Fully automated conversational refund lifecycle via Assistant chatbot — custom NLU model, hybrid dialog system, and deep ecommerce integration eliminating live agent involvement for standard refund cases.
Behavioral segmentation signals enabling tiered support experiences for high-value users — spend-based HVU classification replacing legacy criteria, with Privacy Working Group compliance embedded from design.
Anomaly detection platform and account review tool empowering 200+ fraud analysts — graph-based relationship scoring, ring/chain visualization, and bulk suspension workflows protecting the ad exchange ecosystem at global scale.
For Enterprises
Each project represents an architecture decision that created compounding value. Let's discuss how to apply the same thinking to your platform.
Schedule a Strategy ConsultationFor Founders
These projects were built inside Big Tech. The patterns are portable. Let's talk about what applies to your startup's architecture and what doesn't.
Discuss Your Startup's AI Vision