Career Profile
As a Senior ML Engineer at Altea Healthcare, I lead the development of healthcare AI solutions that directly impact patient care and clinical workflows. I spearhead a team of 6 ML engineers in designing and deploying production-ready AI products including medical information retrieval systems, intelligent transcription tools, and real-time clinical alert systems. My expertise spans the full ML lifecycle from research and architecture design to production deployment, with a focus on building scalable healthcare AI infrastructure using LangGraph, RAG pipelines, and cloud-native technologies. I have established comprehensive evaluation frameworks, observability systems, and CI/CD pipelines on Azure to ensure reliable and compliant healthcare AI deployments. My work combines deep technical knowledge in NLP, vector databases, and MLOps with domain expertise in clinical workflows, enabling the delivery of AI solutions that generate significant value for healthcare organizations. I am passionate about leveraging cutting-edge AI technologies to improve patient outcomes and reduce clinician burden through intelligent automation and decision support systems.
Publications
Submitted: Artificial Intelligence in Medicine (Elsevier), 2024
Published: IEEE Access, 2024
Published: Neural Networks (Elsevier), 2023
Submitted: IEEE Transactions on Image Processing, 2023
Published: IEEE International Symposium on Circuits & Systems (ISCAS), 2023
Submitted: International Symposium on Circuits and Systems (ISCAS), 2021
Published: Machine Learning for Signal Processing (MLSP), 2020
Published: Journal of the International Neuropsychological Society (JINS), 2020
Skills & Proficiency
Python
- Numpy, Pandas, Scikit-learn
Machine learning
- Pytorch, HuggingFace, LangChain, Tensorflow, Keras, Explainable AI (XAI), Bayesian Neural Networks, Healthcare, Bioinformatics, Feature Selection, Image Processing
Platforms
- Git, HPC, GCP, AWS, Azure
Digital Signal Processing
Agile/Scrum
MATLAB
LaTeX
Experiences
- Lead a team of 6 ML engineers in developing and deploying 4 healthcare AI products, driving significant value generation for the organization
- Architected and implemented Aari, a medical information retrieval chatbot integrated into patient charts using LangGraph, with comprehensive evaluation and observability frameworks via LangSmith
- Developed Smart Scribe, an AI-powered transcription system that automatically captures clinical encounters and populates structured notes into predefined sections, improving documentation efficiency
- Built Smart Assessment and Plan Tool that analyzes ICD-10 codes, medications, vitals, and lab data to automatically generate clinical assessments and treatment plans
- Created Condition Detection system that processes nursing notes in real-time to identify critical conditions requiring physician intervention, enhancing patient safety
- Designed and implemented automated RAG pipeline for ingesting external patient documents, including chunking and vector database population for seamless historical data retrieval
- Established end-to-end CI/CD pipelines and cloud infrastructure on Azure for all products, ensuring reliable deployment and scalability
- Drove product strategy and technical architecture decisions while maintaining hands-on development responsibilities across the entire ML product suite
- Wrote Retrieval-Augmented Generation (RAG) Tutorial Paper
- Designed and implemented Moffitt Electronic Health Record Foundation Model to extract and abstract a set of pre-defined descriptive fields and endpoints from Moffitt EHR
- Designed and implemented LLM-based RAG pipeline for sensitive document sanitation
- Designed and implemented VAE to predict anomalies in network traffic
- Designed and implemented feature extraction pipeline to perform Network Address Translation (NAT) detection on real time network traffic
- Collaborated on research studies on deepfake and AI generated text detection
- Implemented cross-attention transformer approach to fuse audio, video and text data to predict clinical endpoints.
- Designed and implemented flexible multi-model feature extraction pipeline using AWS Batch to extract Digital Biomarkers from audio and video data.
- Designed and implemented automated video quality assessment pipeline for data filtering and cleaning prior to use in downstream prediction models.
- Collaborated on research studies with industry partners such as University of Rochester, Bristol Myers Squibb, Boehringer Ingelheim, and others.
- Designed and implemented signal processing pipeline for auditory breathing data
- Designed and implemented feature extraction and model selection pipelines to convert audio to tidal volume and respiratory rate
- Implemented hyperparameter optimization via Bayesian search
- Implemented Influence functions in PyTorch
- Implemented and tested machine learning pipeline for detecting Acute Respiratory Distress (ARD) events in children. Paper has been submitted to Anesthesiology (ASA)
- Designed, implemented and tested early warning system for mortality in ICU based on vital signs
- Served as a consultant to co-workers on deep learning technologies such as PyTorch and PyTorch-lightning
- Researched and tested various deep learning solutions for Radar Discrimination
- Presented findings to colleagues from MIT, SEG, and APL
- Purchased hardware for future machine learning efforts
- Updated Quality Management System and briefed NSWCPD Commanding Officer and Technical Director
- Worked with the Chief Engineer on the standard operating procedure for Risk Management at NSWCPD
- Designed MS Access-SQL Database for metric tracking
Projects
Deployment of an ICU Mortality Prediction Model
- ML backend for my mortality prediction web app. This model has been trained to predict all-cause mortality based on first-day lab values. The model achieves state of the art performance over traditional mortality scores such as APACHE and SAPS. The novelty of the method is in its explanation and uncertainty quantification via influence functions and Bayesian inference.
Influence Pytorch
- Extension of influence function implementation for PyTorch. To support my models, a PyTorch compatible implementation of influence functions was needed. This implementation includes end-to-end gradients, enabling efficient computation of feature attribution.
Reinforcement Learning
- Implementation of Monte Carlo methods to find the optimal blackjack policy. This method achieves the theoretical maximum win-rate without card counting.
Machine Learning
- Implementation of classical machine learning methods such as Linear/Logistic Regression, SVM, MLP, etc