I am a Founding Scientist at Kily.ai where I lead the AI charter. We're building AI for brands and vendors to optimize and grow their sales across online channels (including e-commerce and q-commerce platforms). My work primarily focuses on research and development of small language models, online learning, reinforcement learning, and multi-agent LLM systems.
Previously, I was an Applied Scientist at Amazon, where I led ML initiatives for traffic quality and supply quality programs, pioneering work on scaling GPT-style models for user action prediction and content/data quality assessment on the internet for ad adjacency optimization.
17th AdKDD (KDD 2023 Workshop)
We demonstrate scaling properties across model size, data and compute on User Activity Sequences data. Larger models show better downstream task performance on response prediction and bot detection.
Innovative Applications of AI, AAAI 2023
An approach towards detection of bot traffic that protects advertisers in realtime from online fraud, with an optimization framework for optimal performance across various business slices.
NeurIPS 2022 Workshop on Table Representation Learning
A method for self supervised learning on large scale tabular data, showing efficacy on bot detection with very high cardinality categorical and large range continuous features.
NeurIPS 2021
A speech dataset of over 340,000 spoken words in 50 languages, with over 23.7 million examples.
10th International Workshop on Approaches and Applications of Inductive Programming
We utilize graphical program synthesis to synthesize programs that represent Bongard problems, serving as good representations for learning interpretable discriminative theories using ILP.
Founding Scientist
Building agentic AI systems for e-commerce optimization—online learning, RL, and multi-agent systems
Applied Scientist
Led ML initiatives for traffic quality and supply quality programs, building fraud detection systems at scale.
Data Engineering for Zero/Few Shot Multilingual Keyword Spotting
Work on Optimal LiDAR placement for Autonomous Driving
Pushing the limits of VowpalWabbit with Flatbuffers - achieving 30% performance increase
Particle Track Reconstruction using Machine Learning. 10,000+ particles reconstructed with 93% accuracy in under 10 seconds.
Inductive Programming, Causal ML, and Explainable AI
A PyTorch reinforcement learning library centered around reproducible and generalizable algorithm implementations.
Implementation of Neural Voice Cloning with Few Samples Paper by Baidu Research.
Drone performing imitation learning on IDSIA dataset using Deep Reinforcement Learning.
ML-based reconstruction using Random Forests, XGBoost, Neural Networks, and Graph Neural Networks.
Real-time interfacing capabilities of Conductance based neurons in SpineCreator.
An open source implementation of ChronoNet for seizure detection using EEG data.
