Experience

deepset

• Germany
• Software Development
• 1 - 100 Employee

• Senior Applied NLP Engineer

  • Jun 2022 - Present

  - Leading the scoping and technical execution of proof-of-concept and production-scale NLP solutions on our cloud platform, including retrieval augmented generative AI, and information-retrieval for enterprise clients - Made pull-request contributions to the open-source libraries Transformers, PEFT, and Haystack (70+ contributions) to provide the open source community and our enterprise clients with state-of-the-art NLP solutions - Improved the accuracy of question-answering models by over 10% through gathering and processing large-scale datasets and creating efficient training procedures using Hugging Face's Transformers and PEFT libraries Show less



LF1

• Germany
• IT Services and IT Consulting

• AI Research Scientist

  • Apr 2021 - May 2022

  - Management of partnership with a leading UK‑based eCommerce SaaS company through representation in daily progress and strategy planning meetings, communication and consulting - Leading the development of our retail recommendation system combining deep learning techniques such as NLP and computer vision; Achieved 15‑fold speedups in training and 10% improvement in relevant search results through algorithmic improvements and optimizations (tensor reformulations, mixed‑precision) - Co‑developed, co‑conceptualized, and spear‑headed the testing of the open-source package PADL (Pipeline Abstractions for Deep Learning), which was recognized as one of the 2021 Pytorch Hackathon Winners Show less



Caltech

• United States
• Research Services
• 700 & Above Employee

• PhD Graduate Student

  • Sep 2015 - Nov 2020

  - Derived Lagrangian framework for the projection‑based embedding and the molecular orbital based machine learning models to allow for the calculation of first‑response properties (e.g. nuclear forces) of molecules - Used physically inspired feature engineering to accelerate computational chemistry within the molecular orbital based machine learning framework - Resulted in 7 publications, 3 invited talks and and 10+ poster presentations - Attributed author of three quantum chemistry code bases: QCArchive (Python), QCore (C++), Molpro (Fortran) Show less



The Molecular Sciences Software Institute

• United States
• Software Development
• 1 - 100 Employee

• Project: QCArchive

  • Feb 2019 - Oct 2020

  - Implemented an open-source python interface for Molpro and Entos to execute and process calculations within the QCEngine framework to contribute to the interoperability of quantum chemistry software - Implemented an open-source python interface for Molpro and Entos to execute and process calculations within the QCEngine framework to contribute to the interoperability of quantum chemistry software



Iambic Therapeutics

• United States
• Biotechnology Research
• 1 - 100 Employee

• Project: QCore

  • Jan 2019 - Oct 2020

  - Refactored a C++ implementation of the molecular orbital based machine learning (MOB-ML) method to achieve a O(N^3) scaling reduction - Implemented, code reviewed, and unit tested multiple algorithms used in QCore (a physics based simulation engine) - Refactored a C++ implementation of the molecular orbital based machine learning (MOB-ML) method to achieve a O(N^3) scaling reduction - Implemented, code reviewed, and unit tested multiple algorithms used in QCore (a physics based simulation engine)



Molpro Quantum Chemistry Software

• Software Development

• Project: Molpro

  • Sep 2015 - Jan 2020

  - Contributed over 8000 lines of Fortran code in Molpro, which consists of over a million lines of code - Refactored and modularized the projection-based embedding code to achieve a 20-fold computational speedup, added extensive test coverage and in-depth documentation - Contributed over 8000 lines of Fortran code in Molpro, which consists of over a million lines of code - Refactored and modularized the projection-based embedding code to achieve a 20-fold computational speedup, added extensive test coverage and in-depth documentation