click anywhere
ML Engineer & Researcher · Duke University
Alexander Krol

Designing and building machine learning systems, from research to deployment, across domains.

Duke University ECE + Chemistry + ML GPA 3.739 Graduating 2027
3
Research Labs
4
Publications
3.74
GPA
B.S. + M.Eng.
ECE, Chem, ML
01

Research

Talavage Lab · May 2025 – Present
Medical Imaging ML Pipeline
Machine Learning Engineer
Engineered a transformer-based pipeline to automate wrist bone measurements from X-ray databases, reducing measurement time by 83% vs. manual radiologist workflows. Fine-tuned SAM-Med2D and YOLO on annotated radiograph data, cutting annotation time by 74%. Developed PCA-based geometric feature extraction to classify surgical necessity at ~85% accuracy.
SAM-Med2DVision TransformersYOLOPCAComputer Vision
Kim Lab · April 2025 – Present
Reinforcement Learning & Stochastic Foraging
Reinforcement Learning Researcher
Modeled optimal sequential decision-making under uncertainty as a POMDP-based foraging framework, studying how rational agents develop risk-averse strategies under resource constraints. Investigated correlated random walkers (FFPT) to understand how agent correlation governs collective search efficiency, with direct implications for ensemble diversity and multi-agent RL.
POMDPDDMStochastic ModelingMulti-Agent RLFFPT
Strobbia Lab · May 2025 – Sept. 2025
Nanoparticle Synthesis Automation
Automation Engineer
Repurposed a 3D printer into a programmable syringe pump system with Python automation scripts, achieving a 900% increase in small-scale nanoparticle synthesis throughput. Built an ML classifier predicting nanoparticle morphology across 4 shape classes from reagent concentration inputs. Co-authored SERS manuscript; presented at SciX 2025.
Python AutomationSERSML ClassifierNanoparticlesSciX 2025
02

Featured Project

Cross-Modal Transformer · Jan 2025

Drug-Target Interaction Prediction

Built a cross-modal transformer fusing molecular language (ChemBERTa) with protein sequence embeddings (ESM-2) to predict binding affinity between drug candidates and protein targets. Trained and evaluated on the DAVIS dataset, achieving a concordance index of 0.889, outperforming the DeepDTA baseline. Attention visualization autonomously highlighted known binding sites in 4 of 5 tested drug-protein pairs.

ChemBERTaESM-2Cross-ModalDAVIS DatasetAttention
Model Performance
CI (ours)
0.889
DeepDTA
0.784
Binding sites
4 / 5
ChemBERTa + ESM-2 → cross-attention fusion → binding score
03

Publications

P.01
Normative Bet-Spreading Strategy in an Exhaustible Environment
Krol et al. Derives normative solutions to the optimal stopping problem under diminishing returns, investigating rational risk-taking strategies in resource-limited environments using RL.
Pending
P.02
Correlated Search Strategies and Collective Efficiency in Exhaustible Environments
Kim et al. Examines how correlation among parallel foraging agents affects collective search performance, connecting stochastic walk theory to ensemble diversity in multi-agent RL.
Pending
P.03
Scalable Synthesis and SERS-Based Detection of Nanomaterials
Strobbia et al. Automated pipeline for high-throughput synthesis and SERS detection of engineered nanoparticles.
Pending
P.04
IGF-1R Targeting in Cancer: Does Sub-Cellular Localization Matter?
Soni et al. Journal of Experimental & Clinical Cancer Research, 2023. (Acknowledged)
Published · 2023
04

Technical Skills

Deep Learning
PyTorchTensorFlowJAXHuggingFace
ML Domains
TransformersRLComputer VisionNLPDiffusion
Infrastructure
DockerMLflowFastAPINumPyPandas
Mathematics
Stochastic CalcLinear AlgebraProb TheoryDynamic Programming
Programming
PythonC++JavaSQLMATLABRLaTeX
Wet Lab & Chemistry
Nanoparticle SynthesisSERSqPCRIHCIFWestern BlottingOrganic ChemistrySpectroscopy
Instrumentation
Syringe Pump SystemsX-ray ImagingSignal ProcessingMicroscopy
Languages
English (Native)Spanish (Native)

Available Summer 2026.

Open to research and engineering roles in ML and AI systems. Available starting May 2026.