Open to opportunities

pratik kujur

@pratikkujur

Machine Learning Intern @ Porter | M.Tech25 (Machine Learning) at NIT Karnataka | Developing Production-Grade ML Systems

India

Message

What I'm looking for

I seek roles where I can own end-to-end ML systems and MLOps, work on scalable real-time inference, and drive measurable product impact in collaborative teams.

As a Machine Learning Intern at Porter, I contribute to the development and deployment of production-grade machine learning systems, delivering over 100,000 daily predictions. My work spans end-to-end ML lifecycle tasks, including data ingestion, feature engineering, model training, and real-time inference. I have developed pipelines using Apache Airflow and integrated BigQuery and Snowflake to process millions of records daily while optimizing latency and compute costs.

I have completed M.Tech in Machine Learning at NIT Karnataka, Where I focused on bridging advanced machine learning techniques with practical, scalable solutions. My core skills include Python, LangChain, and Transformers, which I leverage to build impactful systems. Passionate about creating intelligent data-driven solutions, I aim to advance my expertise in applied machine learning and contribute meaningfully to real-world challenges.

Experience

Work history, roles, and key accomplishments

Machine Learning Intern

PORTER

Jan 2025 - Jan 2026 (1 year)

Developed and deployed production-grade ML systems for driver allocation serving 100K+ daily predictions, achieving Precision 0.79 / Recall 0.76 and driving a +0.75pp fulfillment lift; owned end-to-end ML lifecycle and reduced training latency by 90%.

Python XGBoost scikit learn GCP Apache Airflow BigQuery Snowflake MLFlow

Machine Learning Engineer

Porter

Jan 2025 - Jan 2026 (1 year)

Developed and deployed production-grade ML systems for driver allocation serving 100K+ daily predictions, achieving Precision 0.79 / Recall 0.76 and driving a +0.75pp fulfillment lift. Owned end-to-end ML lifecycle and architected ETL pipelines processing 5M+ records/day, reducing training latency by 90%.