Daniel Kaplan Email & Phone Number

New York, Ny, Us

Newcastle Upon Tyne, Gb

All work at Sage done in Python & PytorchImplemented successful graph neural network (GNN) based classifierPrototyped modeling for reading of tables (CSV), using TapasDeployed RAG systems using SoTA generative models, using word embeddings plus LLM based matchingPrototyped various Natural Language to SQL implementations, compared and concentrated with various generative AI solutionsTested and documented the natural limits of LLM bias when solving multiple choice questionsInterfaced with external PhD programs to help guide research goalsAssisted with applied AI course development at universities

Published a paper in ICML 2024, https://arxiv.org/abs/2401.11605.Published a paper in Interspeech 2024.Developed novel research on combining LoRA with generative diffusion models, concurrent with LyCoris.Trained multiple vision languages models, using architectures such as CLIP, SigLIP, CLIPA, DFN, Mistral, Llama, Vicuna, OpenHermes2 Papers in review at Neurips/Neurips Dataset Track 2024.Tags: Computer Vision, Diffusion, Vision-Language Models, Speech Emotion Recognition.

Boston, Ma, Us

Designed and implemented a novel search setup utilizing finetuned CLIP, and internal audio/text data. (Huggingface, Pytorch)Guided/tutored/mentored more junior members of the team and associated teamsBrought additional theoretical discussion to existing problems and solution setsHelped rewrite existed classification search tools for more effective results (Tensorflow)Designed and implemented a joint loss classification system, based on cutting edge Amazon Papers (Pecos, XMC)Created frameworks for rapid implementation and deployment of Seq2Seq systems (Huggingface, Fairseq. Pytorch implementation)

New Brunswick, Nj, Us

Owns the design, development, and implementation of a comprehensive NLP platform, offering a variety of SOTA applications, including model interpretability, Q&A, summarization, translation, and classification.Presented machine learning concepts and processes to leadership, technical and non-technical audiences, to enable J&J Divisions to identify machine learning opportunities.Developed a hierarchical multi-label classification program, using both RNN-LSTM, and transfer learning via BERT.Technical lead: Designed end to end machine learning solutions, and evaluation of implementations. Provided guidance and managed scientists.Interfaced directly with internal clients. Planned and developed customized tools and solutions best suited for customers' particular requirements.

● Presented machine learning concepts and processes to leadership, technical and non-technical audiences, to enable J&J Divisions to identify ML opportunities● Identified over 13 projects that can benefit from ML across the organization● Created an unsupervised Natural Language Processing tool to find matches between products with similar descriptions. This tool was successfully used to match thousands of items located in up to five different systems.● Designed a fuzzy logic tool that leverages Natural Language Processing and Levenshtein Distance in order to find match products, by comparing textual and non-textual attributes.● Implemented a classifier that can predict when truckers will decline to accept a load, an event that increases costs and requires personnel to scramble. Usage of this tool can potentially save millions of dollars.● Developing a system that matches candidates with jobs, by leveraging Natural Language Processing in order to compare work history and job descriptions● Supervised and guided a colleague in designing a regression model that can predict the potential cost of sending items between known and unknown locations

● Designed and developed a machine learning tool to aggregate and analyze real time data from social media in order to identify and track trends. This software is currently being used by the Indonesian Health Ministry to predict the spread of disease, enabling proactive management.● Collected data using Twitter HTTP based RestFul API● Designed a Natural Language Processing classifier to identify relevant tweets using daily data● Design and implemented back end web server using Django and Python● Implemented client side with Javascript, JQuery and HTML/CSS● Developed a heatmap using Google Maps API to display results in an intuitive format● Analyzed results and validated trends against governmental historical data from CDC● Currently testing technology with Indonesian Ministry of Health

• Designed and implemented in Python an image-processing algorithm that generates a times series of activated neurons given a set of images.• Built a custom Peer to Peer web based service to match truckers and clients in need of shipping services. Implemented in Django and Python.

Developed and maintained SOA based software that improves efficiency and reduces costs for the insurance industry• Worked directly with clients using an Agile methodology to implement new features.• Worked under strict time constraints to successfully upgrade large systems to current technologies.

Applied C language to develop a browser based program to capture video and audio from user’s desktops and microphones and convert it to standard video format.Created a browser based video chat program in Java that mirrors Skype applications. Developed a multiplayer turn based game in Java using multithreading and networking.Developed a Japanese to English translation program by applying techniques from natural language processing, compilers, and classification (machine learning), and Lucene.Successfully applied C++ and SQL to create a linking program designed to ease storage and accessibility of information using a combination that is comparable to a Wikipedia page.

Natick, Ma, Us

Successfully revamped the current bug management software to utilize the latest Java technology.Converted SOAP based service into a RESTful (architecture), and created Junit test suite.

Arlington, Va, Us

• Adapted a web based large-scale analytics software product to function on mobile devices.• Implemented in Javascript drag and drop, click, double click• Designed and implemented new front end interface with mobile supported techniques.• Successfully presented final work to senior management and clients who praised and utilized the end product.