👋 I'm Sankalp. I'm currently an undergrad and engineer at the University of Arkansas.
I've created AI-powered prior authorization tools, made contributions to state-of-the-art research in computational neuroscience and quantum materials science, and launched apps that use AI to automate everyday tasks.
l love software that combines research and practical applications. Doing deep dives into generative AI and computer vision right now.
Outside of tech, I like running, playing video games, and trying delicious food.
Work
Resume
Hidalga Technologies
Software Engineering InternSpringdale, AR
Developed a gradient boosting model with 87.6% accuracy on 30K+ cases to predict prior authorization outcomes, reducing human reviews by ≈30%. Piloted an automated document parsing and validation system using Azure OpenAI, tested on 250+ forms and reducing turnaround time by ~60%.
PythonAzure OpenAIMachine LearningDocument ParsingCatBoostPandasMatplotlibLaravelPostgreSQL
University of Arkansas CVIU Lab
Research AssistantFayetteville, AR
Co-authored a physics-informed domain adaptation network (NeurIPS 2025 submission) for aligning synthetic and real 2D material images. Created a synthetic dataset of 600K microscopy images across eight materials and 40 layer types. Improved thickness estimation error by 9.1 nm and detection precision by 30%. Achieved state-of-the-art flake layer classification accuracy of 93.9%.
PythonDeep LearningComputer VisionDomain AdaptationDataset CreationComputational NeuroscienceVision TransformerVariational AutoencoderPyTorch
Projects
High-performance search platform enabling rapid full-text and temporal queries across 50K timestamped webpages. Optimized query processing with vectorized TF-IDF scoring and precomputed indices, reducing latency by 78% and enabling median response times ≤ 400 ms.
SvelteKitAWSNginxPython
Streamlit app automating grocery shopping by converting text recipes or meal images into prefilled Walmart carts. Leveraged ViT, Spoonacular API, and OpenAI GPT for precise ingredient extraction with average processing time ≤ 8 seconds.
StreamlitPythonLangChainViTOpenAI
Pipeline reconstructing images from predicted fMRI signals using Vision Transformers and Variational Autoencoders. Trained on COCO dataset, achieving 33% improvement in reconstruction accuracy.
PythonVision TransformerDeep Learning
Engineered spatio-temporal features from 800K+ SFPD crime reports to classify incidents across 39 categories. Trained a CatBoost model with stratified 5-fold validation achieving weighted F1 score of 0.2505, ranking top 7% on Kaggle leaderboard.
PythonPandasMatplotlibCatBoost
Favorite Games
Baldur's Gate 3
Honor mode run will be complete.
Elden Ring
Beautiful, had to 100% it.
Terraria
Dig, build, fight eldritch monster, repeat.
Sid Meier's Civilization VI
One more turn.
Red Dead Redemption 2
Hey there, Mister!
Ghost of Tsushima
You have no honor. And you are a slave to it.
Sekiro: Shadows Die Twice
Hesitation is Defeat.
Scroll horizontally to see more