• Working on efficient LLM agents for advanced reasoning tasks, optimizing memory and inference workflows.

• Building modular AI pipelines for large-scale experiments and simulations using cloud and containerized infrastructure.

• Contributing to research advancing agentic AI architectures while ensuring scalable and reproducible methodologies.

• Built FastAPI-based backend microservices with Docker, optimizing data ingestion and real-time analytics workflows for scalable deployment.

• Integrated Google AI SDK to design and deploy AI agents for text and vision tasks, fine-tuning LLMs and Vision Transformers for compliant clinical and analytical use cases.

• Developed a full-stack MVP web application, implementing component-based architecture, client-side routing, responsive layouts, and dynamic API integration with backend ML services.

• Orchestrated MLOps pipelines for continuous training, deployment, and monitoring, improving system reliability and model performance.

• Fine-tuned computer vision car classification models using state-of-the-art LLMs and CV architectures, addressing edge cases and increasing production accuracy from 76% → 98%, reducing misclassification rate by 74%.

• Engineered an automated model training pipeline with FastAPI backend, integrated Weights & Biases for model logging, resulting in a 25% reduction in training and processing time, tracking 100+ experiments seamlessly.

• Deployed models for real-time inference via Triton Inference Server, achieving <50ms latency per request and supporting 1000+ concurrent inference requests in production.

• Managed AWS cloud infrastructure, enabling scalable storage, distributed training, and fault-tolerant deployment, reducing downtime by 30% and improving model availability to 99.9%.

• Implemented quantum, quantum-inspired, and hybrid quantum-classical algorithms for large-scale machine learning, optimization, and simulation tasks, achieving up to 1.75× speedup in runtime and 20% improvement in solution accuracy compared to classical approaches.

• Developed and tested algorithms using Qiskit, Cirq, and PennyLane, running 50+ experiments on quantum simulators and near-term hardware, optimizing resource utilization and circuit depth by 15–25%.

• Evaluated algorithm scalability and performance, enabling hybrid pipelines to handle datasets with 10K+ samples and complex optimization problems efficiently.