Sushant Dey

I’m an AI & full-stack engineer who focuses on shipping production-grade systems that feel fast, reliable, and secure. I enjoy designing end-to-end pipelines—from streaming speech signals to LLM reasoning and avatar-based delivery—so the user experience stays crisp under real workloads.

At Orangutan Technologies, I architected a multi-tenant real-time conversational AI avatar platform with stage-isolated WebSocket pooling, achieving sub-200 ms round-trip latency. I also built streaming RAG chat APIs using OpenAI Responses with session-aware context handling, and I’ve worked hard to cut both perceived latency and token spend through smart caching and compressed memory.

I’m equally comfortable with scalable infrastructure and application engineering. I’ve engineered asynchronous ingestion and notification platforms using BullMQ + Redis, implemented Redis-backed caching to reduce redundant LLM/embedding calls, and led an end-to-end DMS platform serving 2,000+ employees with robust, deployable architecture.

I care about secure-by-design access control and operational clarity. I implemented hierarchical RBAC with JWT-scoped middleware to enforce multi-tenant authorization, and I build systems around measurable goals like throughput, SLOs, and reliable delivery semantics.