Aditya Shukla
Machine Learning Scientist
Exploring the frontiers of AI. Designing systems that think clearly, explain themselves, and augment the human spirit... so that they can do our laundry and dishes, and we can do the art and writing.
Research Interests
LLM Reasoning
Inducing robust logical thinking at inference-time
Human-Centered AI
Building solutions that can augment human creativity
Mechanistic Interpretability
Putting models under the mathematical microscope
Explainable AI
Teaching models to answer for themselves
Education
University of Toronto
Master of Science in Applied Computing, Artificial Intelligence
Coursework: Deep Learning - Theory & Data Science, Neural Networks & Deep Learning
SRM Institute of Science & Technology
Bachelor of Technology, Computer Science & Engineering
CGPA: 9.79/10
Professional Experience
Machine Learning Engineer, Pricing & FX
PayPal • Bengaluru, India
LLM-based Reasoning for Contract Intelligence
- Architected a legal reasoning engine with traceable chain-of-thought and clause-level attribution to autonomously analyze PayPal's right-to-reprice merchants across 2,000+ complex pricing agreements
- Designed a modular agent architecture with workflow routing and prompt chaining, benchmarked against 100+ expert-annotated agreements
- Extended pipeline to detect pricing discrepancies, achieving 84% precision and 78% recall, helping claw back $1.5M by flagging undercharging across 50+ legacy contracts
Multi-Agent System for Contract Risk Assessment
- Developed a multi-agent framework using crewAI to analyze business risks across poorly-maintained legacy contracts
- Employed hierarchical task delegation across 5 agents (one orchestrator & four domain experts) with enterprise information retrieval and inter-agent question-answering
FX Trade Volume Forecasting
- Developed a model to forecast PayPal's daily FX trade volumes across 19 currencies to minimize exposure and hedge treasury risk
- Leveraged an ensemble of stochastic (EMA) and deep learning methods (Prophet & Temporal Fusion Transformer) achieving ~30% MAPE on stable currencies
Publications & Patents
Agentic Data Extraction using LLMs
Filed as part of Pricing ML group at PayPal
Patent Pending from US Patent Office
An Integrated Document Manipulation Tool Leveraging LLMs, Transformers, and Attention Mechanism
A. Shukla, P. Ravi, B. Arthi
Best Paper Award at ICIOT'23 • Published in AIP Conference Proceedings 2024
View Publication →Real-Time GPU-Accelerated Driver Assistance System
P. Ravi, A. Shukla, B. Muruganantham
RCAAI'22 • Published in Springer 2023
View Publication →Skills
AI/ML
Deep Learning, LLMs, Explainable Reasoning, Agentic AI
Libraries
PyTorch, transformers, PyTorch Lightning, TensorFlow, Darts, OpenCV, boto3, Flask
Analytics & Data
SQL, GCP BigQuery, statsmodels, R, MongoDB, AWS S3
Cloud/MLOps
AWS CCP, AWS Sagemaker, MLflow, Weights & Biases, Docker, RedHat Linux