Siddhant Sood

Hi, my name is Siddhant Sood and I've worked on a few cool things. Currently spending most of my time doing research @ George Mason University.

Cat
01

Incoming...

At a Forward Deployed Software Engineer at Palantir, I will be deploying AI systems into high-stakes operational environments.

02

Agentic LLM Security Research

At GMU's SECSAT Lab (Security of Emerging Computer Systems and Technologies), researching vulnerabilities in agentic LLM systems, including MCP tool misuse, side-channel leakage, and agent fingerprinting. The work focuses on how open-source agent frameworks expand real-world attack surfaces in deployed AI.

03

Braess' Paradox in Oscillator Networks

At the George Mason Experimental Geometry Lab, derived the first analytical condition predicting when adding an edge causes Braess' Paradox in oscillator networks, and validated it across large-scale simulations. Presented at the Joint Mathematics Meetings 2026.

04

LLM-Driven Energy Grid Simulation

At MITRE, developed LLM-driven simulation agents that continuously updated and executed military energy-grid models for the Office of the Secretary of Defense, enabling live scenario analysis.

05

Semiconductor Microscope Image Alignment

At MITRE's AI & Autonomy Lab, built keypoint-based alignment pipelines for distorted semiconductor microscope imagery using LightGlue and OpenCV, enabling sub-pixel chip inspection,defect analysis and reconstruction.

06

On-Prem LLM Infrastructure

During my time at Blue Sky Innovators, deployed on-prem LLM infrastructure for over 100 employees, reducing inference costs by 80% while improving retrieval precision and hallucination rates for causal AI systems. Also built real-time meeting transcription and semantic search using fine-tuned NVIDIA NeMo and Whisper models.

High School (Class of 2024 @ tjhsst)
07

AI Reddit for Professionals

Started Loops.ai with friends. Built the backend of an AI media-intelligence platform that ingested articles, generated podcasts, and produced personalized research feeds via collaborative filtering. 800 beta testers + 20k invested. YC S'24 interviee + Failed startup lol.

08

CloudGen : Text-to-3D Diffusion Pipeline

Built CloudGen, a diffusion-based text-to-3D pipeline that outputs print-ready CAD meshes and supports bidirectional editing by reversing and rerunning the generative process. This system won 1st Place Grand Prize at Regeneron ISEF 2024 along with other national special awards.

09

Started CerealCodes

Co-founded CerealCodes. Built and ran a global competitive programming platform with 472 teams across 77 countries, managing infrastructure, judging, and outreach. Raised over $40,000 in sponsorships while leading a distributed engineering and operations team.

10

AI-Powered Crisis Hotline

At PennApps (University of Pennsylvania Hackathon), I led a team to build a fully operational AI-powered crisis hotline using OpenAI, Twilio, and reinforcement learning to classify risk and escalate emergencies in real time. The system won 1st Place in AI at the collegiate hackathon (we were in high school).

11

Wearable Sound-to-Nerve Stimulation System

Developed a wearable system that maps sound direction to transcutaneous nerve stimulation using real-time spectrogram models and time-difference-of-arrival localization. The full cyber-physical pipeline ran on embedded hardware and was presented at the IEEE MIT Undergraduate Research Technology Conference.

02

Food Delivery System

At Y Point Analytics, designed and deployed a production multi-agent optimization system coordinating 700+ drivers delivering over 200,000 meals, solving routing and supply allocation under uncertainty. In parallel, started development on an AI therapist copilot that automated clinical documentation and provided continuous patient support.

13

ECG Signal Reconstruction from PPG Data

At Vytal.ai, built deep-learning systems that reconstructed ECG signals from phone-based PPG data, enabling low-cost cardiovascular screening without clinical hardware. This translated biomedical signal processing into deployable ML inside a $12M VC-backed health-tech startup focused on scalable diagnostics.

99% sure it's impossible to score
Beaten: 0 times

Use Arrow Keys or W/S to move your paddle.