Hi, I am Prarthana Krishnamurthy 
I’m a Machine Learning Engineer with a background spanning drug discovery, life sciences, fintech, generative AI, medical imaging, and production ML deployment. Most recently, at Alkermes, I built Quantitative Structure–Activity Relationship (QSAR) models on in-house assay data, then turned them into a FastAPI service, an MCP server that AI agents can call as tools, and a dashboard that three teams use every day. Getting a model into someone’s daily workflow is a different problem than getting it to converge, and it’s the part I enjoy most.
Before that, I spent a year and a half training diffusion and transformer-based generative models for my master’s thesis at Northeastern, running experiments across a cluster of H100 GPUs. The most useful thing I took away was learning how to debug and scale generative models when the usual tricks stop working. I started my career at Société Générale, building ETL pipelines and Airflow workflows that financial systems depend on. That grounding in ML production infrastructure is why my models hold up once they leave a Jupyter notebook.
On the modeling side I work with PyTorch, diffusion models, and GNNs. On the deployment side I use Docker, Kubernetes, AWS, Azure, and Terraform, with Spark and Airflow handling the data itself. I’m drawn to problems where the system either works in the real world or it doesn’t, and where getting it there takes genuine engineering depth.
Outside of work: several hundred hours in Factorio and counting. (Yes, it’s also a production pipeline problem.)
Python
Bash
Linux
Git
NumPy
SciPy
Pandas
CUDA
React
Hugging Face Transformers
spaCy
PyTorch
scikit-learn
TensorFlow
Keras
MLflow
Weights & Biases
FastAPI
Docker
Kubernetes (AKS, EKS)
AWS
CI/CD (GitLab, Jenkins)
Airflow
PySpark
Pytest
Terraform
Kafka
Apache Flink
Hadoop
Elasticsearch
Built the predictive modeling and scoring stack for a research program, spanning data consolidation, model development, and deployed services
Architected transformer and diffusion-based generative models using flow matching for non-autoregressive text generation, scaling training across 8x H100 GPUs with PyTorch FSDP and NCCL at 95%+ GPU utilization. Cut validation loss by 92.3% (0.93 BERT similarity) and reduced DPM-Solver++ discretization error to 0.65.
Deployed object detection labs on AKS and EKS with Docker, backed by Jenkins/Terraform MLOps pipelines and CloudWatch/Prometheus monitoring. Recognized as top-performing TA for modernizing the MGEN 7220 cloud course architecture.
M.S. in Computer Software Engineering with a focus on deep learning applications in healthcare and molecular modeling. Engaged in cutting-edge research exploring the intersection of diffusion models, graph neural networks, and large language models to solve complex problems in medical imaging and drug discovery. Coursework includes Advanced Machine Learning, Deep Learning Systems, Computer Vision, and Natural Language Processing.
Productionized card fraud detection models for a 500K-transaction-per-day portfolio, from feature engineering and model selection through Kubernetes serving and drift monitoring. Architected AWS data lake supporting 1M+ daily records, designing ingestion and partitioning strategies for high-throughput SaaS data processing and compliance reporting. Refactored 50+ legacy scripts into Airflow DAGs and productionized 3+ ML Dockerized apps on AKS handling 1M+ daily financial transactions.
Delivered a retail banking churn proof of concept spanning SQL analysis of 250K customers, calibrated modeling, and a stakeholder dashboard.
Developed and optimized C code for real-time image processing algorithms on Texas Instruments C66x DSP platform. Achieved 16.4-microsecond execution time for block processing through advanced optimization techniques. Reduced cycle count to 4096 cycles per block, achieving 4× speedup through SIMD vectorization and loop unrolling. Implemented efficient memory access patterns and cache optimization strategies for embedded vision applications.
Earned Bachelor's degree in Electronics and Communication Engineering with specialization in neural networks, artificial intelligence, and digital image processing. Developed strong foundation through coursework in Machine Learning, Digital Signal Processing, Computer Vision, and Embedded Systems. Undergraduate research and projects in adaptive image processing using GoogleNet architecture and CLAHE (Contrast Limited Adaptive Histogram Equalization) laid the groundwork for current specialization in deep learning and computer vision. Completed senior thesis on adaptive image enhancement techniques for medical imaging applications.