Yash Singhal

AI Engineer specializing in production LLM systems, retrieval pipelines, and inference optimization. Built AI platforms processing 185M+ patent vectors and 1M+ documents while reducing inference cost by 35% and latency by 40%. Experienced in shipping scalable AI infrastructure on Kubernetes using Docker, Qdrant, and vLLM.

Artificial Intelligence Engineer - Researchwire Knowledge Solutions Pvt. Ltd. - Noida, India

(2026-07)

• Engineered an enterprise-grade RAG pipeline (Docker + Qdrant) for large-scale patent analysis, architecting a distributed system to process 1M+ documents with sub-second semantic search latency across the global corpus.
• Architected custom chunking and embedding strategies that optimized retrieval precision by ~40%, effectively minimizing signal-to-noise ratios in LLM context windows and improving grounding for complex legal reasoning.
• Deployed a high-concurrency vLLM inference server, scaling throughput by ~2–3× while reducing unit cost by 35%; utilized continuous batching and prompt optimization to slash P95 latency from 3s to 1.8s.
• Spearheaded the vectorization of 185M patent records using Turboquant, ensuring high availability and horizontal scalability for massive-scale vector similarity search.
• Implemented a hybrid ML + LLM orchestration layer utilizing lightweight classifiers to pre-filter workloads, reducing LLM invocation frequency by ~50% and significantly optimizing infrastructure compute expenditure.

AI Engineer Intern - Researchwire Knowledge Solutions Pvt. Ltd. - Noida, India

(2026-01 - 2026-07)

B.Tech - Computer Science & Engineering - Jaypee Institute of Information Technology (2026-05)