Applied ML engineer · Bioinformatics
I build, evaluate, and deploy machine-learning products, from scientific prototypes to tested APIs and live cloud applications.
Selected work · 2024—2026
Five projects showing the path from research question to evaluated model to shipped software.
A full-stack force-spectroscopy platform with a built-in interactive demo, FastAPI and Streamlit applications, authenticated MCP tools, and a bounded autonomous agent that combines live scan data with local RAG. Deployed with Docker and automated delivery to AWS.
Python · FastAPI · MCP · Agents SDK · RAG · AWS
A balance-aware approximation for leakage-aware dataset splitting. It cuts the optimization burden while preserving split quality, evaluated across four molecular ML benchmarks.
Python · RDKit · scikit-learn · ILP
Fine-tuned speech transformers to identify 22 Indian languages in a low-resource setting, then traced speaker-dependent failure modes through per-speaker evaluation and embedding analysis.
PyTorch · Transformers · MMS · XLS-R
The classification workstream of a three-person demand modelling project: an imbalance-aware ensemble, explainability across every model, and structured analysis of ordinal errors.
CatBoost · XGBoost · Random Forest · SHAP
A reproducible comparison of segmentation approaches with quantitative evaluation and qualitative error inspection across challenging image regions.
Python · scikit-image · OpenCV
About · The connective tissue
My background sits where data complexity meets strong computational systems. I care about what a model learns, where it fails, and whether someone else can reliably run it.
That means clean experiments, honest evaluation, useful interfaces, and deployment practices that move a project beyond the notebook. I’m looking for a role where I can build reliable systems with measurable real-world value.
Trajectory
Scaling leakage-aware dataset splitting for large ML datasets under Prof. Dr. Olga Kalinina.
Advanced study across machine learning, computation, and the life sciences.
Built a churn-prediction pipeline on 10,000+ transactions and connected model output to stock planning.