Karan Aggarwal

AI Researcher

Seattle, Washington, United States12 yrs 4 mos experience

Most Likely To SwitchHighly Stable

Key Highlights

Stackforce AI infers this person is a Data Scientist with expertise in NLP and Machine Learning in SaaS and Fintech.

Natural Language Processing (nlp)Machine LearningData Science

AlgorithmsC++Computer ScienceData MiningDeep LearningEclipseGANsHadoopHealthcareInformation RetrievalJavaLaTeXProgrammingPythonR

Working with Amazon as an Applied Scientist. Primary interests include time-series analysis and natural language processing.

Sep 2024 – Present · 1 yr 6 mos · Seattle, Washington, United States · Hybrid

Working on context management for code development agents for the AWS's coding assistance platform, Amazon Q Developer.
Lead the science development of /doc, an automated documentation generation feature during Re:Invent 2024.

context managementautomated documentation generationNatural Language Processing (NLP)Machine Learning

2 roles

Promoted

Oct 2023 – Present · 2 yrs 5 mos · Seattle, Washington, United States

Apr 2020 – Sep 2023 · 3 yrs 5 mos · Seattle, Washington, United States

Working in Amazon Fintech providing solutions for information extraction from documents and emails for efficient payment systems and cash optimization. Previously, I worked on a confidential project in Amazon Devices.
Main problems I've worked on: weak supervision, semi-supervision, and synthetic data generation using pre-trained large language models (LLMs).

Natural Language Processing (NLP)Time Series AnalysisMachine Learning

Jun 2019 – Aug 2019 · 2 mos · Adobe Research, San Jose

Worked on a method to cluster discrete time event sequences in a evolutionary way to model user behavior. Proposed a time varying clustering method to model user generated sequences. This work was published at SIGIR 2020.

clusteringuser behavior modelingData Science

Feb 2019 – May 2019 · 3 mos · Criteo AI Lab, Palo Alto

1) Build domain adaptation models for cold starting product sellers on user prospecting platform by using re-targeting data. We propose two supervised domain adaptation methods that achieve superior performance over the baselines. This work has been accepted as a long paper at RecSys 2019.
2) Showed that GANs can be used for regression problems with a capability of generating complex noise patterns including heteroscedastic and multi-reponse noise forms.

domain adaptationGANsregression problemsMachine Learning

Jun 2018 – Aug 2018 · 2 mos · Hitachi Industrial AI Labs, Santa Clara

Worked on deep learning and survival analysis methods for combining time-to-failure and failure event probability prediction for equipment monitoring using sensor data. This work was published as a long research track paper at IEEE BigData 2018.

deep learningsurvival analysisMachine Learning