Hi, I'm Omkar Nayak.
A Scientist
Self-driven, quick starter, passionate programmer with a curious mind who enjoys solving complex and challenging real-world problems.
Mathematics and Computer Science Student at the University of Michigan - Ann Arbor. Machine Learning Engineer at Synafox AI and Research Associate II at Zhou Lab.
Currently developing autonomous agentic workflows for clean energy infrastructure and working on publications in LLMs and GWAS.
Projects
Multi-agent system using CrewAI for monitoring and optimizing clean energy infrastructure through digital twin simulations.
Novel phenotype recognition method using LLMs for Genome Wide Association Studies with 50,000+ genetic records.
Integrating expert medical reasoning into LLMs, improving diagnostic accuracy by 38% on 80,000+ health records.
Analysis of the Monitoring the Future dataset using data analytics tools to discover trends in attributes of adolescents in the US and predicting their political learning.
Intelligent news aggregator using NLP and machine learning to classify, summarize, and recommend trending articles from multiple sources.
Experience
- Designed and developed autonomous agentic workflows using CrewAI for the continuous monitoring and optimization of digital twin simulations improving the management of predictive models for system analysis and resource allocation.
- Integrated of multiple Agentic Frameworks into the CI/CD pipeline, automating the software development cycle by using MLOps principles to minimize manual intervention and accelerate the Agile iteration frequency.
- Deployed scalable, production-grade RAG pipelines using LangChain and vector databases to serve domain-specific LLMs. This architecture ensured low-latency, context-aware inference for specialized enterprise applications.
- Tools: Python, CrewAI, LangChain
- Enhanced Genome Wide Association Studies (GWAS) at the UM School of Public Health by authoring a paper on a novel phenotype recognition method, the bespoke Large Language Models (LLMs) successfully tested on 12 distinct phenotypes from over +50,000 Genetic Sequencing and Electronic Health Records.
- Advanced Deep Learning diagnostic accuracy by about 38% by collaborating on a paper within a cross-functional team of doctors at Michigan Medicine to integrate expert reasoning into LLMs. This work analyzed +80,000 precision health records from +24,000 patients, using NeMo Guardrails to ensure data security and topic relevance.
- Accelerated AI-driven research cycles by 15x on a HIPAA-regulated HPC cluster by implementing a parallelized software architecture with vLLM within an Agile development framework while utilizing Site Reliability Engineering techniques to employ information security methods to protect sensitive data.
- Tools: Python, LangChain, LangGraph, PubMed Central API, PyTorch, vLLM, Bash
- Deployed Predictive Modeling using Lightwood and Hugging Face models on Docker to develop 5+ proprietary ML Models.
- Analyzed information using the internal database and integrated ML models into the SAAS platform services of ”Suuchi GRID”, resulting in a 27% increase in orders.
- Tools: Python, MindsDB, Hugging Face, Lightwood, PyTorch, Apache Superset, Docker, PostgreSQL
Education
University of Michigan - Ann Arbor
Ann Arbor, Michigan, USA
Degree: Bachelor of Science in Computer Science and Mathematics
- Graduate Machine Learning
- Graduate ML and SP in Biomedical Sciences
- Graduate Probability Theory
- Machine Learning Research
- Computer Vision
- Advanced Data Science
- Data Structures and Algorithms
Relevant Courseworks:
Skills
Frameworks
CrewAI
LangGraph
LangChain
Keras
TensorFlow
PyTorch
Flask
Django
CrewAI
LangGraph
LangChain
Keras
Libraries
SciPy
MLflow
OpenCV
scikit-learn
Pandas
NumPy
Matplotlib
SciPy
MLflow
OpenCV
scikit-learn
Developer Tools
Python
Git
AWS
PostgreSQL
MATLAB
Shell Scripting
HTML/CSS/JS
MindsDB
Hugging Face
Python
Git
AWS