Aryan Sinha

Built an agentic financial research platform using MCP orchestration and LangGraph to enable LLMs to query live market data and validate research via Vector-RAG, reducing hallucinations through cross-referencing across 50+ 10-K filings.

Architected a distributed computer vision Manager–Worker platform to autoscale CV pipelines across 50+ edge nodes, reducing idle compute costs by 25% and enabling programmatic pod provisioning via a FastAPI control plane interacting with Kubernetes.

Deployed and scaled multimodal inference pipelines across GPU clusters, reducing onboarding time by 40% and achieving mobile OCR latency of 1.2s using vLLM KV caching and a custom Kotlin SDK. Engineered a gRPC microservices mesh and fault-tolerant Kubernetes platform to sustain 500+ TPS and maintain 99%+ SLO compliance.

Designed a high-throughput video indexing pipeline ingesting RTSP streams via Kafka and serving Qwen2-VL workers with vLLM, enabling sub-200ms semantic retrieval over 2M+ embeddings in Milvus.

Optimized real-time detection pipelines with TensorRT and FP16 quantization to increase throughput by 10 FPS and improved spatial query efficiency by 70% via MongoDB-based zone mapping for ReID systems.

Completed a Bachelor of Technology in Electronics & Telecommunication with coursework and projects focused on systems, real-time video analytics, and distributed inference pipelines.