Aarushi Sinha sinhaaarushi

Systems focused engineer working on real time infrastructure, simulation, and LLM safety pipelines

Designing systems that don't break under load routing, backpressure, real time state
Building simulation first tools to reason about system behaviour instead of guessing
Exploring LLM pipelines under adversarial conditions (injection, leakage, noisy inputs)
Prioritising observability and reproducibility over surface level features

Systems > features. Focus on behaviour, trade offs, and failure modes.


Problem	Simulating real time task routing decisions under load with measurable trade offs
Solution	Tick-based simulator with WebSocket live feed, metrics, replay, seeded RNG, twin strategy compare, JSON export.
Stack	Node, Express, `ws`, React, Vite, TypeScript, Vitest, GitHub Actions


Problem	Simulating autonomous driving requires coordinating perception, decision making, and control across multiple models
Solution	Built a multi model pipeline combining computer vision and decision logic to simulate real time driving behavior
Stack	Python, computer vision, deep learning, simulation pipeline


Problem	Product support systems struggle to retrieve accurate, context aware information from large knowledge bases
Solution	Implemented a RAG pipeline using retrieval + generation to answer queries with improved relevance and context
Stack	Python, OpenSearch, HuggingFace, retrieval pipelines


Problem	Coordinating multiple agents for decision making and task handling lacks clear modeling and evaluation under different conditions
Solution	Built an agent based simulation system to model interactions, decision flows, and outcomes across multiple agents in a controlled environment
Stack	Python, agent logic, simulation workflows, backend processing

Increasing realism and control in task routing simulation (latency, queue behavior)
Strengthening AUTODRIVE decision pipeline across perception → action flow
Studying system behavior under failure and uneven load conditions

Provide feedback