Designed and optimized the RAG retrieval stack using hybrid search and cross-encoder reranking and implemented an efficient context construction strategy that drastically reduced hallucinations.

Engineered financial-domain LLMs by applying supervised fine-tuning (SFT) and parameter-efficient fine-tuning (PEFT) methods to open-source models, achieving improved performance on internal benchmarks while reducing computational overhead and latency.

Developed software and algorithms for an AI-powered 510k robotic medical device, NovaGuide 2 for blood flow analysis, stroke, and heart shunt grade assessments.

Designed and implemented AI-based anomaly detection, real-time medical data visualization, robotic probe control, and other core features that significantly accelerated diagnostic workflows and enhanced clinical decision making by providing clearer, more actionable insights.

Engineered a 50% reduction in deep learning computer vision model training time by implementing distributed training on a multi-GPU high-performance computing system.

Developed and deployed an anomaly detection model on an edge device to identify anomalous artifacts in radar data in real-time.

Prototyped formal verification methods for Artificial Neural Networks (ANNs) in autonomous flying vehicles to address challenges in AI-driven, safety-critical control systems.

Engineered high-performance computing solutions by parallelizing optimization problem computations with CUDA for Julia, addressing demanding mathematical workloads.