M.S. Computer Science · UW–Madison
Maithil
Mehta.
Building at the intersection of Systems, AI, and Full-Stack engineering.
About
01 / 04I'm an M.S. Computer Science student at the University of Wisconsin–Madison (GPA 4.0), previously a Senior Member Technical at D. E. Shaw Group where I built large-scale data infrastructure and analytics platforms across three years.
My interests span distributed systems, machine learning, and full-stack engineering. I hold a B.E. in Computer Science with a Minor in Data Science from BITS Pilani, and I enjoy bridging deep systems knowledge with applied AI to ship things that scale.
University of Wisconsin–Madison
M.S. Computer Science
Sep 2025 – May 2027 · Machine Learning, Foundational Models, Distributed Systems, Big Data
Teaching Assistant — Data Management for Data Science
BITS Pilani
B.E. Computer Science · Minor in Data Science
Aug 2018 – May 2022 · Data Structures & Algorithms, Foundations of Data Science, Information Retrieval, Graph Mining
Languages
Machine Learning & Data Science
Generative AI & LLMs
Big Data & Distributed Systems
DevOps & Tools
Software Development Practices
Experience
02 / 04MathWorks
Software Engineering Intern
May 2026 – Aug 2026
Natick, MA
- Built an autonomous LLM agent for a 50TB infrastructure data warehouse, using MCP tool-calling and schema-driven APIs to translate natural language into validated queries across 45+ endpoints.
- Engineered the agent's reasoning framework for multi-step planning, parameter extraction, and sequential/parallel API workflows, with caching, validation, and enterprise SSO for safe data access.
The D. E. Shaw Group
Senior Member Technical — Investor Relations
Jul 2022 – Jun 2025
Hyderabad, India
- Architected a data platform with temporal modeling (Python, SQL) over 100M+ records, delivering a 2.5× growth in active platform usage through optimized query pipelines and automated validation workflows.
- Shipped a React full-stack Language-to-Visualization tool used by 100+ stakeholders, enabling natural language-driven analytical plot generation and eliminating dependency on analyst intermediaries for ad-hoc reporting.
- Reduced data validation turnaround from days to hours by engineering automated anomaly detection and reconciliation pipelines across multiple upstream data feeds.
Amazon.com, Inc.
Software Development Engineering Intern — Finance Automation
Feb 2022 – Jun 2022
Hyderabad, India
- Reduced manual finance review effort by 35% by building JavaScript and Ruby on Rails tooling that automatically surfaced edge-case discrepancies in vendor invoice workflows across high-volume transaction data.
- Cut operational maintenance costs by 25% by leading migration of critical finance APIs from on-premises Java infrastructure to native AWS, improving reliability and reducing deployment overhead.
Salesforce, Inc.
Software Engineering Intern — Big Data Cloud
Jun 2021 – Jul 2021
Hyderabad, India
- Slashed incident resolution time by 30% by building a multithreaded Python library for real-time deployment metric tracking across distributed data centers, with integrated alerting and anomaly detection.
- Won the company-wide intern hackathon (competing against 200+ interns) by engineering a computer-vision-based 2FA liveness verification system using OpenCV and TensorFlow.
Projects
03 / 04ALPS Serverless Scheduler via Google ghOSt
Custom implementation of the ALPS (Adaptive Learning, Priority Scheduler) policy using the Google ghOSt userspace framework on CloudLab (Intel Xeon Gold). Mixed-workload microbenchmarking revealed a critical 'blind spot' where I/O-bound tasks induce starvation for CPU-bound functions, increasing their tail latency by 15% vs. Linux CFS baseline.
eBPF-Driven Analysis of Linux CFS
Low-latency kernel instrumentation using eBPF/bpftrace to directly trace pick_next_task_fair() in the Linux Completely Fair Scheduler. Validated CFS proportional fairness by measuring vruntime differentials across three priority levels, confirming that the highest-priority process (nice -15) received the longest time slices.
TALMAS: Training-Free Attention Suppression for Diffusion LMs
Invented a training-free inference intervention for Masked Diffusion Models that suppresses [MASK] token noise via dynamic, asymmetric logit biasing, improving generation coherence on GSM8K and HumanEval with negligible O(n) overhead and no loss in parallel decoding speed.
Course-Specific Exam Assistant — Fine-Tuning vs. RAG Pipeline
Built and benchmarked a hybrid RAG pipeline (Elasticsearch for lexical retrieval + ChromaDB for semantic search) against a LoRA fine-tuned Llama-3.2 model as a TA-led exam prep tool for 150 students. Quantitative analysis showed hybrid RAG reduced hallucinations by a statistically significant margin for transcript-based Q&A, informing the deployment decision.
CV-Based 2FA System — Salesforce Hackathon Winner
Won the Salesforce Intern Hackathon by building a computer vision-based two-factor authentication system in Python using OpenCV and TensorFlow, eliminating reliance on external hardware or mobile tokens. Engineered a facial recognition module with reflected light-based liveness detection to mitigate deepfake and photo spoofing attacks.
CNN Leaf Disease Detection with CLAHE
Enhanced plant leaf disease classification by applying CLAHE on the luminance channel (OpenCV) for better feature extraction. Trained Keras models on the augmented PlantVillage dataset (38 classes), benchmarking against a modified VGG-16 with CLAHE and segmentation. Achieved 98.5% accuracy, comparable to AlexNet.
Fake News Detection via Geometric Deep Learning
Graph Neural Network to classify news as fake or real based on its propagation structure in a social network, independent of URL content. Demonstrated that propagation-based approaches complement content-based methods, enabling accurate detection after ~7 hours of spread. Achieved 91.7% ROC AUC on the Kaggle Twitter dataset.
Personal Portfolio Website
High-performance portfolio built with Next.js 15 App Router, TypeScript, and Tailwind CSS. Features a filterable project gallery aggregating work across Systems, AI/ML, and Full-Stack domains, with Framer Motion animations and a dark-first design system.
Contact
04 / 04Let's build something.
Open to research collaborations, internships, and full-time opportunities. Reach out any time.