San Francisco Bay Area
I’m interested in advancing machine learning systems by translating research ideas into scalable solutions, with a focus on multimodal learning and understanding how models, data, and infrastructure generalize across diverse domains! My recent work spans multimodal systems, including low-latency XR applications with on-device inference, constrained LLM reasoning frameworks, and production RAG pipelines for financial document analysis with an emphasis on evaluation, robustness, and deployment at scale.
AdTech, MarTech
Commander’s Intent & Course-of-Action Synthesis (Office of Naval Research) - Building a multi-agent reasoning and information retrieval system for strategic military narrative generation -Increased intent accuracy 30% and reduced manual corrections 40% by building an LLM-guided decision system combining PEFT-tuned transformers with symbolic planning – Designed statistical evaluation metrics for closed-loop model behavior, improving system reliability in autonomous decision settings – Productionized the full pipeline in C++ and Python for real-time decision environments
– Designing and implementing a next-generation Energy Management System with Siemens Energy, processing 500K+ real-time status signals and 250K analog measurements, showcasing innovative distributed computing techniques for high scalability and reliability. – Engineering a distributed optimization solution using MPI and the Julia/JuMP framework for power grid operations on HPC nodes, achieving computation speedup and advancing predictive modeling at scale.
- Engineered and deployed an end-to-end low-latency on-device ML modules for AR/XR, integrating Micro XR frameworks to deliver real-time (sub-15 ms) object and gesture detection for the XR AI Solutions Team - Designing and developing text-to-music and text-to-sound compact and efficient models for nearly real time music and sound generation
- Achieved 94% accuracy in architecting a production RAG system to assess financial document quality and relevance at scale, supporting semantic search, summarization, and downstream decision-making - Collaborated across 4 business units to develop and scale a SQL-backed Resource Data Service (RDS) using Java SpringBoot, integrating data from 10+ internal trading systems catering to 10+ downstream requirements through high-performance Kafka-driven APIs to deliver real-time analytics
-Drove the development of chaos engineering protocols within the Aurora Storage Fabric team, generating real-time data analytics from 1M+ simulation events, which informed fallback strategies for large scale event failures. – Established a one-touch automation pipeline using Java Guice framework, identifying critical architectural gaps and rewriting the metric validation framework improving effiecency by 35%