M.S. Computer Science · AI/ML · UMD

Mehak
Gupta

Software engineer & ML researcher building intelligent systems — from neural networks to production-grade applications.

View My Work Contact Me
✉ Email LinkedIn GitHub
Degree
B.S. + M.S. CS
Focus
AI / ML
GPA
3.8 / 4.0
Graduation
2026
MG
Mehak Gupta
Software Engineer & ML Researcher

CS student at the University of Maryland pursuing a combined B.S./M.S. with a concentration in AI/ML and a Minor in Statistics. I love working at the intersection of machine learning research and software engineering — fine-tuning LLMs, building full-stack apps, or designing ML pipelines from scratch.

PythonJavaC/C++ ReactNode.jsMongoDB PyTorchTensorFlow OpenCVDockerSQL RustOCaml
Technical Skills
Frontend & Full-Stack
Backend Engineering
Machine Learning / Deep Learning
Systems & Low-Level (C, Rust, OCaml)
Data Analysis & Statistics
Languages
PythonJavaTypeScript C/C++RustOCaml RubyMATLABSQL
Featured Projects

Things I've Built

[ secure ATM system ]
Secure ATM-Bank Communication
C · OpenSSL — May 2025
Secure protocol between ATM and bank server assuming a fully compromised router. AES-128-CBC and HMAC-SHA-256 with replay, tampering, and brute-force protections.
COpenSSLCryptography
GitHub
[ budgeting app ]
Budgeting App
HTML · CSS · JavaScript — May 2025
Mobile budgeting prototype with income tracking, spending analysis, and goal setting. HCI principles via user research, personas, and iterative accessibility testing.
JavaScriptHTML/CSSHCI
GitHub
[ git assistant ]
Command Line Git Assistant
Python · NLP — Dec 2024
CLI tool for Git. Dataset generated with small LMs; fine-tuned BART-Base to produce accurate, context-aware Git commands from natural language input.
PythonNLPBART
GitHub
[ LLM image agent ]
LLM Agent for AI-Generated Images
Python · TensorFlow — Dec 2023
Framework addressing limitations of text-to-image diffusion models by optimizing prompts and evaluating outputs through iterative LLM feedback loops.
PythonTensorFlowDiffusion
GitHub
[ micro-ocaml compiler ]
Micro-OCaml Compiler
OCaml — Apr 2023
Full compiler pipeline: lexer, parser building an AST, and interpreter executing MicroOCaml expressions with functions, commands, and recursion.
OCamlCompilersAST
GitHub
[ unix shell ]
Custom Unix Shell
C — Dec 2022
Unix-style shell built from scratch: file redirection, piping, AND chaining, and subshells with full process management.
CUnixSystems
GitHub
[ youtube geolocation ]
Transcript-Based Geolocation of YouTube Videos
Python · PostgreSQL/PostGIS — Feb 2026
Developing a transcript-based geolocation system that uses NLP, geocoding services, and PostgreSQL/PostGIS spatial queries to infer real-world video locations from noisy metadata and transcript clues.
PythonPostgreSQLGIS
GitHub
Experience

Where I've Worked

Software Engineering Intern
Synapse Labs
Jul – Aug 2024
  • Fine-tuned prompts for GPT API calls, improving response accuracy and efficiency.
  • Researched and implemented alternatives to CTGAN — including Wasserstein networks and Gaussian copula models — improving synthetic data quality by 30%.
Data Analyst
Unity Group
Jul – Aug 2023
  • Analyzed stock market trends and built algorithms to determine optimal stock exit timing, maximizing user profits.
  • Researched economic indicators and industry factors to design effective investment strategies.
Software Development Intern
SAIL — Steel Authority of India Limited
Jun – Jul 2023
  • Designed an order processing and inventory system with Python and MySQL, cutting manual processing time by 40%.
  • Built Tkinter GUI and MySQL integration for seamless, efficient data handling.
Teaching Assistant — OOP Java I & II, Computer Systems
University of Maryland
Aug 2022 – Present
  • Led discussions for 36 students on data structures, algorithms, and programming in Java and C.
  • Graded exams, quizzes, and projects; held office hours to support student learning.
Research

What I'm Investigating

🖼️
Automated Scientific Poster Evaluation
NLP · Multimodal · VLM
Developed standardized evaluation metrics for poster generation systems beyond inconsistent VLM-as-a-judge scoring — incorporating content quality, design quality, and layout coherence. Ran human studies to analyze alignment, reliability, and evaluator disagreement.
🫁
Speech Detection from Physiological Signals
Signal Processing · ML · Dr. Ashok Agrawala
Investigating whether speech can be reliably detected from respiratory and accelerometer signals without a microphone. Building signal-processing pipelines and ML models to capture subtle physiological patterns differentiating speech from non-speech.
🧠
Implicit Bias in Over-Parameterized Deep Networks
ML Theory · Deep Learning · Optimization
Investigated how optimization hyperparameters influence implicit regularization, double descent, and generalization. Designed controlled experiments analyzing learning rate, batch size, momentum, and early stopping under modern theoretical frameworks.
📡
WiFi-Based Indoor Navigation System
Signal Processing · Indoor Localization
Developing a real-time indoor navigation system using smartphone WiFi scans and multi-access-point signal triangulation to track user movement across floors and rooms in GPS-denied environments.
🌍
Multilingual Theorem Proving Analysis
NLP · Formal Methods
Investigating why multilingual training improves automated theorem proving through controlled experiments comparing monolingual and multilingual ProofWala models under matched training data size, focusing on tactic diversity as a mechanism for stronger pass@k performance.
Contact

Let's Connect

Whether it's a research opportunity, internship, collaboration, or just a conversation — I'd love to hear from you.

✉ mgupta25@terpmail.umd.edu 💼 LinkedIn ⌨ GitHub 📞 +1 240-726-2922