Curriculum vitae
I work where rigorous mathematics meets systems that can be built, tested, and verified—from generative image models to formalized analysis and economic forecasting.
dmongia@berkeley.edu (661) 383-6159 LinkedInResearch braid
Choose a discipline to follow it through the chronology.
Foundation
Theory first. Application always in view.
Expected May 2027
Evidence ledger
Five records · September 2024 to August 2026
Showing all 5 records.
Trained a three-step image-generation model that maps random noise to an image in one network call, improving FID from 83.65 to 13.17.
Built from the drifting architecture introduced in February 2026, the model removes the need for a pretrained encoder and generates interpretable CIFAR-10-derived samples at evaluation.
Tools: PyTorch, capped-gradient corrections, clean-fid, torchvision, FID/KID, precision/recall, CUDA, NumPy, Matplotlib, and Git.
Worked with undergraduate researchers and machine-learning professors to study and implement drifting-based image-generation methods.
Investigated the mathematical foundations of generative models and translated theoretical ideas from recent papers into working implementations and experiments.
Methods: PyTorch, JAX, GPU computing, FID, and KID for model development, experimentation, and evaluation.
Reframed core ideas from real analysis through filters, the abstraction used in Lean 4 to formalize limits and continuity.
Developed a research paper centered on clear technical exposition, mathematical arguments, and standard academic writing practices.
Tools: Lean 4 for definitions, theorem statements, and proofs; Mathlib for verified mathematics; LaTeX for writing and styling; and Git.
Completed an applied-mathematics research project and paper over one semester, then presented the findings to Berkeley mathematics faculty.
Developed experience interpreting academic literature, structuring a paper, presenting mathematical arguments, and communicating technical results to peers and mentors.
Evaluated several machine-learning models for U.S. GDP forecasting using historical macroeconomic data and economic indicators.
Published a step-by-step technical article explaining the machine learning, mathematics, and macroeconomic methodology behind the project.
Tools: Python for modeling, sourcing, and cleaning; Pandas for data organization; and Matplotlib and Seaborn for model evaluation, analysis, and visualization.
Working toolkit
Methods over buzzwords.
Python and Pandas for implementation, sourcing, cleaning, and organization.
Experiment design, accelerated training, numerical work, and evaluation.
Machine-checked definitions and proofs paired with technical writing.
Reproducible workflows, AI-assisted implementation, and visual explanation.
Next problem
I’m interested in opportunities involving machine learning, finance, economics, and applied mathematics.
CV source updated August 2026 · Berkeley, California