Inquisitive Innovator Award
Recognition for research and product development.
Research & engineering
Siddharth Magesh
M.S. Computer Science, University of Southern California
Language Model Reasoning & Verification · Multimodal Learning · AI Systems
I’m a master’s student in Computer Science at the University of Southern California, working at the intersection of language-model verification, multimodal learning, and agentic systems. My current research examines autoregressive–diffusion complementarity, verifier error correlation, and calibration under constrained inference budgets. Through GRIDS, I’m also developing permission mechanisms that reconcile adaptive agent behavior with explicit user-defined boundaries.
Previously, at IIT Madras’s Sudha Gopalakrishnan Brain Centre, I adapted language and vision-language models for computational neuroscience and digital histology. My engineering experience spans resource-constrained perception, hierarchical agent orchestration, reusable Python infrastructure, and production backend architecture, connecting model-level experimentation with dependable system design.
Awards & Recognition
Research & Current Work
Verification Methodologies for Reliable Mathematical Reasoning
A comparative study of outcome, process, and generative verification.
My current technical review examines how outcome, process, and generative verifiers assess mathematical reasoning, comparing supervision, reliability, and inference cost. It proposes a compute-controlled evaluation and informs my ongoing investigation of autoregressive–diffusion verification.
OSSARTH
An LLM-native operating system research prototype.
A prototype exploring language-model-driven task planning and orchestration over a Linux-based system. My work included hierarchical agent orchestration, MCP tool integration, and a simulated task-management interface. I co-authored a research manuscript submitted for publication.
Selected Projects & Open Source
Sentimatrix
Multimodal sentiment intelligence, packaged for developers.
A Python package for asynchronous data collection, sentiment and emotion analysis, and multimodal inputs. I built cloud/local LLM integration, CLI and REST API interfaces, and structured exports for analysis and reporting. The project connects reusable analysis workflows with practical developer interfaces.
Deep Learning Sandbox
Architecture implementations and reproducible experimentation in PyTorch.
A research and implementation workspace spanning CLIP, GPT-2, residual networks, Siamese networks, and vision transformers. Combines architecture notes, notebooks, and modular PyTorch experiments.
VEC Website Backend
Production backend engineering and team leadership.
Backend architecture and implementation for Velammal Engineering College’s website, alongside team coordination, deployment, documentation, and handover through WebOps.
Experience
Graduates Rising in Data Science (GRIDS)
Oct 2026–PresentMember · Los Angeles, California
Developing Permissions That Learn: fixed user-defined boundaries, a decision gatekeeper, and approval-learning memory for AI agents. Evaluating interruptions, unauthorized actions, and permission creep in simulated AgentDojo tasks.
Sudha Gopalakrishnan Brain Centre, IIT Madras
Dec 2025–Feb 2026LLM Researcher · Chennai, India
Adapted LLMs and vision-language models for computational neuroscience and digital histology. Built pipelines for Nissl/H&E imagery, MRI-linked data, and biomedical annotations, covering alignment, fine-tuning, inference, and evaluation on NVIDIA DGX A100 infrastructure.
Explore the work →WebOps Club, Velammal Engineering College
Feb 2025–Feb 2026Vice Chairman & Lead Backend Developer · Chennai, India
Helped lead a 17-member team building the college’s production website. Architected Node.js/Express/MongoDB services and 50+ REST endpoints, with authentication, AWS deployment, CI/CD, monitoring, and security controls; established reviews, documentation, and successor mentoring.
Explore the work →Youth Red Cross, Velammal Engineering College
Nov 2022–Oct 2025Technical leadership & community service · Chennai, India
Progressed through three roles spanning community programs, volunteer coordination, and software development.
Technical Head
Aug 2024–Oct 2025Led technical tools and automation, including a blood-donation platform connecting donors with people in urgent need.
Joint Secretary
Sep 2023–Aug 2024Coordinated event planning, volunteers, logistics, scheduling, and on-ground delivery.
Volunteer
Nov 2022–Sep 2023Participated in 15+ community initiatives, including donation drives, awareness campaigns, and outreach.
Inzpire Solutions
Feb–May 2025Software Developer Intern · Tamil Nadu, India
Developed ML stock-transfer prediction and verification, utensil-cleanliness classification, and inventory vision workflows for retail and Smart Café operations. Built end-to-end datasets and model pipelines; published clean/dirty glass classification, object detection, and Rangoli Detection datasets on Kaggle.
Explore the work →Scrapify Ecotech Private Limited
Nov 2024–Jan 2025AI Software Engineer · Tamil Nadu, India
Developed EcoFloater’s onboard AI for aquatic-waste detection and navigation. Integrated LLM decision components, YOLOv8, DeepLabV3+, lightweight vision transformers, and sensor inputs on Jetson Nano; applied compression, quantization, and latency optimization for local mission execution.
Explore the work →Licenses & Certifications
Let’s connect
I’m happy to connect about language-model reasoning, verification, multimodal learning, and AI systems. Email is the best way to reach me.
smagesh@usc.edu