Sanat Sharma Email and Phone Number

Built a Multimodal AI understanding (using a novel supervised contrastive loss) system trained on Adobe data to understand user creative intent. The model is used in several features including search and recommendations of over 400million stock assets.Working on pre training and fine-tuning multimodal LLMs. Built a RAG system for Adobe content retrieval. The work was presented at AAAI 2024.Built a user Query to Adobe Product matching system that improved product CTR significantly and reduced null rate by nearly 50%. The model is now used in Creative Cloud Desktop, Home and other surfaces. The work was presented at SIGIR 2024.Worked on a system to augment knowledge graphs by automating hierarchy generation using LLMs. The hierarchies are used in several production features including search and SEO. The work was presented at ECIR 2024.Working on SLMs and LLMs for search, including Multimodal LLMs, pretraining, finetuning and optimization.Overall works have led to 11 patent filings and 5 conference papers (SIGIR, CVPR, AAAI, ECIR)

Built the Creative Knowledge graph, a graph to bridge user creative intent to useful assets and recommendations. The graph is used to power several Adobe use case such as Seo page creation, intent detection, recommendations.Created a model for Font Recommendations in Adobe Express (recommending relevant fonts based on user text), used by millions of monthly users.Created Layout Bert, a model for document and image layout understanding, which broke the state-of-the-art for several layout datasets. The work was presented at CVPR AI4CC 2022 (https://arxiv.org/abs/2205.00347)

Created a multilingual neural spellchecker for search queries. The model provided significantly higher accuracy on English and French queries (than traditional spellers like Hunspell), and is used in production to power spell correction across various Adobe surfaces. The work was presented at SIGIR 2023.

2Learn (https://2learn.in) provides curated content to engineering and science university students in India. Launched during the pandemic in July 2020, 2Learn brings over 3000 hours of university-specific content to help students pass and ace their University exams. We currently support students from over 20 colleges.

UT Longhorn Entrepreneurship Agency (www.utlea.org) is dedicated to fostering and supporting entrepreneurship at the University of Texas at Austin. As the hub for entrepreneurship on the 40 Acres, we work with the university's administration and student orgs to cultivate a premier campus environment for students to build new ventures. Some of the events put on LEA include EWeek (Entrepreneurship Week), Freshman Founders, Founders table and more.EWeek 2020, held in Feb 2020 attracted over 2000 students to 20+ events throughout the week, from tech talks with the VP of DoorDash to VC panels.

Represent University of Texas Computer Science to prospective students, and give frequent tours to prospective students, parents and guests.

2WayLive is an interactive in-classroom Tele-education platform, aimed at connecting great teachers to students around the world (https://2WayLive.com ). 2WayLive aims to disrupt the in-classroom learning industry in India, by allowing teachers to broadcast live lectures to students in an efficient, cost-effective manner.Provided live classroom connectivity to a large network of over 100 cities.Texas Shark Tank - 2nd placeTexas Venture Labs Pitch - 3rd Place

Worked on the Language Understanding team, responsible for providing machine learning models to do semantic and language understanding tasks for MS assistant (Cortana), Microsoft Teams, Outlook etc.Developed a classifier to detect “outlier” user queries (queries Cortana is inept at) and constructed a training pipeline to train the LU models on these outliers, in order to make them more robust. Model was able to generate a stratified sample of outlier queries and perform significantly better on them.Worked with various word embeddings (BERT, GloVe), deep learning models(Attention-based LSTMs, BERT) and feature extraction/generation techniques.

TA for CS 303E, Introduction to Computer Science for non-majors

Worked on the DOM team in the Edge browser. Worked on the creation of a garbage collected caching mechanism that would be used throughout the browser to cache strings allocated in the browser, thus helping boost speed and memory performance, helping save upto 5mb of memory on certain websites.Also worked on the getAttributeNames api which is a W3C mandated api which will now be supported on Edge!

CS Roadshow is an outreach organization that focuses on spreading awareness about and teaching STEM and Computer Science in underprivileged schools (middle-high schools). If you're interested in having us come to your school, contact us at utcsroadshow@gmail.com or shoot me a message!

Worked on creating one of the earliest compilers from TypeScript to WebAssembly, a binary encoding format for the web. WebAssembly is compatible with most web browsers and has the potential of changing the way highly computational processes on the web take place.Worked with the TypeScript team as well as an external project, AssemblyScript, to develop the compiler.The initial prototype got over 2000 stars on Github.

Worked on ‘Middleware SOA architecture’ for a large British bank. o Design aspects included creating service layout using UML, IBM’s Rational Software Architect and Soap UI. o Build aspects included providing security and service‐functionality to a number of the bank’s operations. Worked on IBM Websphere DataPower and MessageBroker platforms to create functioning services. Gained insight into data analysis, security and back‐end orchestration with databases

Worked on the creation of a data entry business that would employ special needs(specifically Autism) and homeless people. Tied up with Texas Parent to Parent, a non-profit to service clients. Oversaw all the technical research and aspects of the project, as well as interviewed a lot of industry specialists to test viability.The company was a student based venture aimed at making a difference and giving back to the community.

Worked on a research project on Robotic Automation, with special focus on its impact on the BPO sector. Did extensive research on the impact of automation for public good and its implementation in fields ranging from agriculture, healthcare to manufacturing. The Research Paper was well received with special appreciation from the Vice President of the company.

Worked on a research project on Cloud Computing, with special focus on its impact in the education sector. Researched on the different forms of Cloud Computing, with special emphasis on SaaS and its implementation in the satellite based teaching sector.