cv

AI Research Engineer

Subconscious.AI

Benchmarked SGThread system and worked to deploy novel inference engine and RL trained model to serverless platform.

• Containerized and modified fork of SGLang, facilitating rapid deployment of TIM reasoning model.
• Created python interface for Google DataCommons benchmarking of modified inference engine and TIM model through the creation of constrained decoding specifications and a new DC LLM tool.

Backend/Kernel Developer

BitEnergyAI

Developed code for FP8 E4M3 multiplication on FPGA in C++ with Vitis-AI HLS to integrate transformer architecture with efficient floating point arithmetic operations.

• Designed host interface for multiplication between NxD weight matrix and DxM input matrix.
• Implemented FP8 multiplication and addition through fused multiply add in C++.
• Optimized matrix multiplication kernel through Vitis High-Level Synthesis best design practices.
• Conducted software and hardware emulation of FPGA design.

Full-stack/NLP AI Engineer Intern

SALIERI.AI

Developed a server-less SaaS deployment with an autoscaling Kubernetes cluster hosted on GCP to facilitate structuring LLM output into parsable JSON based on business text document stores.

• Utilized Node.js to create a web-hook API, enabling communication with a Python "child process" via inter-process communication (IPC).
• Developed 3 containerized micro-services for scalability and created network for intra-container communication.
• Restricted LLM output using Context-Free Grammars in GBNF format and JSON to ensure valid generation (token generation constraint).
• Built and optimized an autoscaling Kubernetes cluster with MySQL data store on Google Cloud Platform, with containers stored in Google Artifact Registry and configured using Helm Chart for long-term sustainability.

Founder

Uplift.co 501(c)(3)

Organized first cohort and facilitated weekly meetings to grow the project to a peak of 30 members.

• Assisted in the creation of 5 in person wellness events and 10 digital wellness campaigns.
• Built components of brand including website, Instagram, and initial digital wellness campaigns for public awareness of organization.

Candidate for Bachelor of Science

Massachusetts Institute of Technology (MIT)

Major 6-3 Computer Science and Engineering

• Deep Learning
• Mathematics for Computer Science
• Data Structures and Algorithms
• Linear Algebra
• Machine Learning
• Probability & Statistics

MIT Student Information and Processing Board (SIPB) (Keyholder)

MIT

Harvard x MIT CO-OP (Student Board Member)

Harvard University & MIT

International Science and Engineering Fair (Finalist)

Society for Science

THREAD, "Thinking Deeper with Recursive Spawning"

NAACL 2025

Phillip Schroeder, Nathaniel W. Morgan, Hongyin Luo, James R. Glass. THREAD, "Thinking Deeper with Recursive Spawning" [NAACL 2025] (arXiv preprint arXiv:2405.17402 2024)

ARK (Automated Resource Knowledge Base)

Leading a team of 6 across MIT, GT and UIC to create a community LLM app store and modular personalized agentic assistant development platform.

• Configured LLM Inference with Hugging-Face TGI and Qwen-2.5 weights hosted on HuggingFace Hub.
• Created 'base_module interface for agentic tool calling and inference engine abstraction layer.
• Organized project outline including "GitHub Org.", task list, weekly meetings, and various repositories for maintaining current and future contribution consistency as project scales.
• Secured $5000 grant from MIT SIPB for development server build.
• Built and set up headless dev server including installing CUDA kernel, configuring Linux distro, adding user SSH keys, and configuring user permissions.