Aakash Varma Nadimpalli

Applied Scientist · Amazon · Seattle, Washington
nadimpalliaakashvarma@gmail.com

I am an Applied Scientist at Amazon working on large-scale machine learning and Generative AI systems. My work spans LLMs, NLP, transformer architectures, supervised fine-tuning, parameter-efficient adaptation, reinforcement learning, RAG, and scalable ML systems. I develop and productionize machine learning systems that operate at large scale, including LLM-driven data pipelines, product-fit prediction, model evaluation, hallucination and faithfulness safeguards, distributed training, and cloud-scale inference. Previously, I worked as a Machine Learning Researcher at Enelect, where I developed deep learning and computer vision systems for real-world applications. I hold a PhD in Computer Science from Wichita State University, with research focused on deepfake detection, domain adaptation, generative AI, and machine learning generalization.

Experience

Applied Scientist

Amazon
  • Led development of large-scale LLM and Spark pipelines that transform unstructured customer feedback, including reviews and returns, into structured product-fit signals consumed by production retail systems.
  • Built end-to-end ML systems to infer garment and footwear fit across apparel, shoes, and sports categories, generating millions of product-level fit labels for downstream retail experiences.
  • Designed NLP extraction models with negation-aware parsing to identify directional fit signals from customer reviews and accurately interpret nuanced customer language.
  • Developed cohort-normalized scoring methods to correct base-rate skew and improve prediction accuracy and generalization across diverse product populations.
  • Designed and trained transformer-based cross-attention models for size and fit prediction using customer purchase and return histories; explored two-tower and set-pooling architectures to improve cross-category and cold-start generalization.
  • Owned the transformer training lifecycle from tokenization and distributed training through checkpointing, ONNX export, and offline evaluation, delivering scalable, production-ready models.
  • Fine-tuned task-specific language models using supervised fine-tuning (SFT) and parameter-efficient methods including LoRA and QLoRA, optimizing model quality while reducing training and inference resource requirements.
  • Applied reinforcement learning and preference-based optimization to improve LLM behavior and task performance; evaluated smaller language models (SLMs) against larger frontier models and optimized SLMs for competitive quality with lower inference cost and latency.
  • Architected hybrid rule-based and LLM systems for automated product content generation, including deterministic faithfulness and hallucination guards and LLM-as-Judge evaluation to validate factuality, helpfulness, and quality.
  • Built and optimized AWS-based ETL and inference pipelines using Spark, Athena, S3, SageMaker, and Bedrock, processing tens of millions of records per run; implemented data-quality validation, temporal holdouts, CI/CD, monitoring, and regression detection for production reliability.
Aug 2025 - Present

Machine Learning Researcher (Computer Vision)

Enelect
  • Led development of deep learning and computer vision models for image segmentation, classification, anomaly detection, and real-world visual intelligence applications.
  • Designed semi-supervised learning (SSL) algorithms and novel loss functions to improve annotation quality, model generalization, and robustness on large-scale datasets with limited labeled data.
  • Integrated Generative AI techniques, including diffusion models and adversarial training, to improve data efficiency and robustness for computer vision systems.
  • Built cloud-native deepfake detection systems on GCP and AWS, enabling scalable, low-latency inference and production deployment.
  • Engineered batch and real-time ML pipelines using Airflow, Kafka, and Spark; developed multimodal embedding and temporal-modeling systems for coordinated manipulation, identity spoofing, and anomaly detection.
  • Built a domain-specific agentic RAG system using LangChain, Milvus, quantized 8B LLMs, and vLLM, achieving 92%+ accuracy and ~200 ms latency for real-time Q&A over research documents.
  • Applied production MLOps practices including CI/CD, training-serving skew minimization, A/B testing, drift monitoring, and automated evaluation to maintain reliable, version-controlled ML deployments.
Jan 2023 - Aug 2025

Graduate Research Assistant

Wichita State University
  • Developed scalable and transferable deep learning methods for domain adaptation and facial-manipulation detection, leveraging generative AI and unlabeled data to improve robustness across diverse datasets.
  • Developed generalizable deepfake detection models targeting deployment across social media platforms, strengthening AI-driven security and authentication.
  • Published and presented research at leading AI conferences on deepfake detection, AI security, fairness, and explainable machine learning.
Aug 2020 - Dec 2022

Research Intern (Data Management)

ZOLOZ Inc.
  • Planned computer vision data-collection efforts supporting biometric authentication and spoof-detection research.
  • Built and executed data-collection workflows to generate diverse, high-quality datasets and classify attributes for downstream model training.
  • Developed spoof-attack methods and synthetic spoof data to improve adversarial robustness and detection accuracy.
Jan 2020 - Jun 2020

Technical Skills

Languages: Python, SQL, C++, CUDA; familiar with R
Machine Learning: PyTorch, TensorFlow, Keras, Scikit-learn, PySpark, Deep Learning, NLP, Computer Vision, Transformer Models, Statistical Modeling, Anomaly Detection
LLM / Generative AI: LLMs, SFT, LoRA, QLoRA, Reinforcement Learning, Preference Optimization, RAG, LangChain, Hugging Face, vLLM, GPT, Transformers, LLM-as-Judge, Prompt Engineering, Diffusion Models, GANs
ML Systems / Data: Apache Spark, Spark SQL, Airflow, Kafka, ETL, Batch and Real-time Pipelines, Distributed Training, Model Evaluation, Experimentation
Cloud / MLOps: AWS (SageMaker, Bedrock, S3, Athena, EC2), GCP, Docker, Kubernetes, GitHub Actions, CI/CD, ML Monitoring, Weights & Biases, TensorBoard
Libraries: NumPy, Pandas, SciPy, OpenCV, NLTK, spaCy, Matplotlib

Education

Wichita State University

PhD in Computer Science
Wichita, Kansas

Thesis: Passive and ProActive Methods for Facial Forgery-based DeepFake Detection

Aug 2020 - May 2024

University of Missouri-Kansas City

MS in Electrical Engineering
Kansas City, Missouri
Aug 2018 - May 2020

Jawaharlal Nehru Technological University Hyderabad

Bachelor's Degree in Electronics and Communications Engineering
Hyderabad, India
Sep 2014 - May 2018

Publications

Harnessing Unlabeled Data to Improve Generalization of Biometric Gender and Age Classifiers

Authors: Aakash Varma Nadimpalli , Narsi Reddy , Sreeraj Ramachandran , Ajita Rattani
Abstract
IEEE Symposium Series on Computational Intelligence (SSCI), 2021

An Experimental Evaluation on Deepfake Detection using Deep Face Recognition

Authors: Sreeraj Ramachandran , Aakash Varma Nadimpalli , Ajita Rattani
Abstract
International Carnahan Conference on Security Technology (ICCST), 2021

On Improving Cross-dataset Generalization of Deepfake Detectors

Authors: Aakash Varma Nadimpalli , Ajita Rattani
Paper
IEEE/CVF CVPR Workshop, 2022

GBDF: Gender Balanced DeepFake Dataset Towards Fair DeepFake Detection

Authors: Aakash Varma Nadimpalli , Ajita Rattani
Abstract
International Conference on Pattern Recognition (ICPR), 2022

Demographic Fairness and Accountability of Audio and Video based Unimodal and Bi-modal Deepfake Detectors

Authors: Vinaya Sree Katamneni , Aakash Varma Nadimpalli , Ajita Rattani
Abstract
Face Recognition Across the Imaging Spectrum (FRAIS), 2023

ProActive DeepFake Detection using GAN-based Visible Watermarking

Authors: Aakash Varma Nadimpalli , Ajita Rattani
Paper
ACM Transactions on Multimedia Computing, Communications, and Applications, 2023

Facial Forgery-based Deepfake Detection using Fine-Grained Features

Authors: Aakash Varma Nadimpalli , Ajita Rattani
Abstract
International Conference on Machine Learning and Applications (ICMLA), 2023

Social Media Authentication and Combating Deepfakes using Semi-fragile Invisible Image Watermarking

Authors: Aakash Varma Nadimpalli , Ajita Rattani
Abstract
ACM Digital Threats: Research and Practice, 2024

Recommended Reading

Beyond LoRA: Can You Beat the Most Popular Fine-Tuning Technique?

Hugging Face · June 2026

A practical exploration of parameter-efficient fine-tuning (PEFT) beyond LoRA, examining alternative adaptation methods and how to choose the right technique for fine-tuning open language models under real-world compute and memory constraints.

Advanced Fine-Tuning Methods on Amazon SageMaker AI

AWS Machine Learning Blog · July 2025

An overview of modern LLM fine-tuning and alignment techniques, including LoRA, QLoRA, PEFT, RLHF, and DPO, with an emphasis on selecting approaches based on model size, compute constraints, and production requirements.

Fine-Tuning Your First Large Language Model with PyTorch and Hugging Face

Hugging Face · February 2025

A hands-on walkthrough of quantized model loading, LoRA, supervised fine-tuning (SFT), and Hugging Face TRL. The article provides a practical end-to-end view of adapting an open language model to a specialized task.

We Benchmarked 12 Small Language Models Across 8 Tasks to Find the Best Base Model for Fine-Tuning

Distil Labs · December 2025

An empirical look at small language models (SLMs) and how fine-tuning can close the gap between compact models and much larger models, highlighting the importance of model selection, fine-tuning strategy, and task-specific evaluation.

Interests

Outside of my work in machine learning and artificial intelligence, I enjoy traveling and outdoor activities such as hiking and camping.

I am interested in astrophysics and follow research and developments related to celestial bodies and space.

I also enjoy exploring different cuisines, watching and playing cricket, and going to the movies with friends.

Awards & Honors

Wichita State University

Global Select Scholarship Award (GSSA)
Aug 2020 - May 2024

University of Missouri-Kansas City

Dean's International Scholarship Award (DISA)
Aug 2018 - May 2020

Review Activity

Reviewer for ACM Transactions on Information Forensics & Security, Image and Vision Computing, IEEE Transactions on Circuits and Systems for Video Technology, Pattern Recognition Letters, International Conference on Pattern Recognition (ICPR), and IET Biometrics.

```