Devarsh Kale

I’m a machine learning engineer intern at Vosyn.ai, building models for real-time voice localization and translation for products like VosynCore and VosynConnect.

I designed, implemented, and trained ML models, running experiments to improve accuracy and inference efficiency by refining algorithms and tracking results to support better decision-making.

My experience spans both computer vision and NLP: I built ML pipelines for medical image classification (X-ray/MRI) to reduce false positives and improve accuracy by ~15%, and I developed an NLP system to simplify medical discharge instructions, reducing reading complexity by ~40–50% using structured prompt engineering.

I also develop production-style prototypes, including a Streamlit + Flask application with LLM-based validation, PDF upload, and export, and I’ve built an end-to-end churn prediction pipeline on 10k+ records using feature engineering, SMOTE balancing, and model evaluation (Random Forest achieving ~85% recall and ~0.85 F1-score) with batch inference via joblib. I enjoy moving from experimentation to deployable systems using tools like REST APIs and Docker.