Open to opportunities

Prasanth kumar Duddu

@prasanthkumarduddu

Message

Associate Data Scientist specializing in production AI for payments and fraud detection.

India

Message

What I'm looking for

I seek roles where I can build and productionize scalable ML/LLM systems, improve model performance, and collaborate with cross-functional teams to secure high-volume payments.

I am an Associate Data Scientist with 3+ years of experience building production-grade AI for India's payments ecosystem, focused on fraud detection, LLMs, and scalable data pipelines.

At NPCI I led TB-scale data curation for FiMI, optimized vLLM inference to reduce latency and increase throughput, and engineered reusable evaluation frameworks and preprocessing workflows that accelerated iteration cycles.

I built Autoencoder-based UPI fraud models, engineered 30+ behavioral features, automated SQL-driven pipelines, and helped productionize systems serving millions of daily transactions while improving recall and reducing false positives.

Experience

Work history, roles, and key accomplishments

Current

Associate Data Scientist

Current

National Payments Corporation of India

Jan 2023 - Present (3 years 2 months)

Built and productionized AI systems for India's payments ecosystem, leading TB-scale data curation for a sovereign LLM and deploying optimized vLLM inference that reduced latency 35–50% and increased throughput 2–3x; developed Autoencoder-based UPI fraud models that improved recall 12–20% while lowering false positives 10–20%.

Python SQL LLM VLLM Data Curation Docker Model Evaluation

Machine Learning Intern

Feynn Labs

Nov 2022 - Dec 2022 (1 month)

Built deep learning architectures (ANN, CNN, RNN) including LSTM/GRU for sequential tasks and applied CNNs for image analysis; supported experimentation using A/B and statistical hypothesis testing.

TensorFlow Keras CNN RNN A B Testing Hypothesis Testing Python

Machine Learning Intern

AryaDrj Technologies

Aug 2022 - Oct 2022 (2 months)

Prepared datasets and engineered features, applied scaling and optimization techniques, and implemented clustering and ensemble methods to improve model performance.

Data Cleaning K Means Python scikit learn & Model Tuning