Suraj Vathsa Email & Phone Number

San Francisco, Ca, Us

San Mateo, California, United States

Worked on several high-ownership projects such as: - AI Search and Alerts: Worked on developing an industry-first AI-powered natural language search and alerts product that allows users to semantically search security footage or proactively receive alerts using natural language descriptions leveraging large foundation models. Further implemented an end-to-end system to provide users with suggestions for natural language queries specific to their organization and suggest semantically similar searches at query time.- People and Vehicle Analytics: Worked on improving the object tracking based people and vehicle analytics pipeline. Implemented and tuned a novel night-time people detection algorithm to improve recall by ~20% on an end-to-end testing dataset. Developed a novel technique to run the people and vehicle analytics on warped fisheye cameras by making use of on-camera hardware accelerated de-warp APIs and coordinate translation to run multi-object tracking in a warped space.- Real-time License Plate Recognition: Worked on implementing a real-time license plate recognition software that runs on the edge at 20 FPS using the byte-track multi-object tracking algorithm. The system showed ~10% improvement in both precision and recall. Involved in development in all ends of the stack including training OCR models, setting up testing infrastructure, backend api server and frontend fixes to help launch a new License Plate Management user interface.

Worked on using AI-based CV approaches to solving vehicle attributes search and license plate recognition

Los Angeles, California, United States

Intern at house calls and telehealth start-up Heal.• Worked on developing an in-house system to process insurance cards using Deep Learning during patient on-boarding.• Used Recurrent LSTM networks to identify the patient name and member ID and a Convolutional Neural Network to identify payer name on top of AWS rekognition character rekognition service. Obtained a 98+% test accuracy on the member ID classification and payer name detection tasks.• Explored the use of Graph Convolutional Neural Networks for classifying the entities in the insurance card and thus extracting information from the same.• Deployed the system as a scalable serverless Lambda function for consumption by client-side services.• Worked on developing a basic prototype in the iOS client to consume the insurance card lambda service.• Engaged in a literature review for ICD-10 based disease trajectory prediction.

ACM ICPC is a group dedicated to improving the algorithmic thinking and implementation skills for students interested in pursuing technical interviews. Organized workshops for the “Interview Track” at ACM ICPC. Taught concepts on various Algorithms and Data Structures and helped prepare students for technical interviews in Computer Science.

New York, New York, United States

Intern at a fast paced AI startup that uses NLP techniques to provide customer feedback insights.• Worked on expanding and scaling the data-ingestion pipeline to ingest over 50000 customer interaction points from public sources like Facebook, Instagram, and Re:amaze.• Worked on optimizing dynamic ingestion to update data sources regularly with new customer data using a cron job.• Worked on network graph optimizations to improve the interpretabilty in the graph by deduplicating small reviews and removing islands formed in the graph.

Los Angeles, California, United States

Machine Learning Research Intern at the Scalable Analytics Institute (ScAi) at UCLA. Working on developing a Neural-Symbolic approach to solving the problem of Multi-Hop Question Answering. Investigated the use of state of the art recurrent evidence retriever and developed symbolic operators. Helped develop a multi-task deep learning model to help with operator and evidence span selection which can be utilized to extract useful entities and construct answers.

San Jose, California

Extended a health monitoring service to monitor the health of infrastructure microservices like Elasticsearch, Kafka and Zookeeper in a containerized application orchestrated using Kubernetes. The service was designed to monitor heartbeat pings from services and track vital metrics using EWMA.Additionally, developed an autohealing framework for the Elasticsearch microservice to solve the problem of exceeding disk space watermark levels.