Wael Salloum, Ph.D.

Mendel is a clinical AI platform for life sciences and healthcare organizations that deciphers health data with clinician-like logic. It indexes structured and unstructured data and replaces the complexity of querying with the simplicity of chatting. My research group works on hybrid AI (ML and symbolic) with novel approaches to: - Large Language Models - Graph Neural Networks - Knowledge representation: going beyond ontologies to a better conceptual understanding - NLP/NLU on clinical text (EMR and literature) - Natural language questioning of clinical knowledge - Reasoning... Lots of reasoning... and clinical common sense - Knowledge acquisition, discovery, and consolidation - And some computer vision sprinkled on top

As a founding member (the first employee) of EMR.AI Inc., a healthtech AI startup transforming clinical documentation using Artificial Intelligence, I built the NLP department, technologies and code base from scratch, and helped hire and train fellow scientist. EMR.AI’s Medical Dictation software allows physicians to dictate episodes which are then formatted into reports and EMR records. I developed the language modeling component of our Automatic Speech Recognition (ASR) system that we showed to be on par with professional human transcribers. I also built various NLP components responsible for formatting the text (dates, numbers, etc.), predicting punctuations, predicting report structure (e.g., paragraphs, section headers, and numbered lists), identifying preamble text and extracting meta-data from it, among others. I also built EMR.AI’s Automatic Coding software that reads free text reports and identifies ICD-10 codes which help speed up the billing process. I also designed a prototype for an Automatic Scribe, a software that listens to doctor-patient conversations and produces a summary (reports, EMR entries, outpatient letter, etc.) based on the current and previous interaction, and patient’s medical history. Most of the research I done at EMR.AI is published and I filed multiple US patents.

As the chief scientist of MedTools, I developed a scalable, semantic search engine on medical device information with instant-search capabilities. I also built OntoLogic, a system that facilitate the rapid creation of ontologies by learning from text and having human experts teach the algorithm by correcting its errors. I worked with a physician to build our own medical device ontology using OntoLogic. I implemented a reasoner (inference engine) that uses the ontology to enrich the semantic search engine. Also, I built a search query generation system (provides autocomplete and spelling error correction via domain specific language models and context models), and a recommendation engine that uses a mix of ontology-based and collaborative filtering algorithms. As a CTO, I managed a team of a dozen or so developers building different physician-facing services such as a Medical Note Generator, a Preference Sheet collaborative building and management center, a medical device unboxing video center, a mobile app, a CPT code search engine, among others.

I did my Ph.D., M.S., and M.Phil. degrees at Columbia University, during which I worked as a Graduate Research Assistant on Machine Translation from Arabic Dialects to English. Funding and Projects: - 2010-2011: Funded by DARPA's GALE project. - 2011-2015: Funded by DARPA's BOLT project. - 2015: Funded by Google. Some of my research work at Columbia was packaged and released in two systems to be used by DARPA GALE & BOLT teams: ADAM and ELISSA. I also helped build another resource: Tharwa. In addition to GALE and BOLT, we participated in two Machine Translation competitions: OpenMT Eval 2012 and 2015. We ranked second in OpenMT Eval 2015. We also participated in EMNLP-2014 QALB Shared Task for Error Correction and won first place.

Worked on proposing a new model for Semantic Question Answering systems that resulted in a paper published in 2009.