Omkar Manjrekar Email & Phone Number

Mountain View, California, United States

Mountain View, California, United States

Production Engineer for Search

I am a Teaching Assistant to the graduate ML class at Stony Brook University taught by Prof. Yifan Sun. My responsibilities involve helping students with their questions about the assignments, project or lectures, and grading homework.

Stony Brook, New York, United States

Courses: Machine Learning, Computer Vision, Natural Language Processing, Robot Learning, Distributed Systems, Probability and Statistics

Menlo Park, California, United States

I worked as a SWE intern at Meta in the summer of 2022. I was a part of one of the feed content classification ML teams at Meta which classified facebook and instagram content (posts, stories, reels, etc.) into one of the several categories. Very briefly, I wrote data pipelines, a lot of SQL, configured dashboards for showing results and authored ML training workflows. • Created an embedding monitoring dashboard (using an internal tool like Grafana) for continuous monitoring of the performance of the content classifiers. Additionally, the dashboard allowed us to systematically understand the data and uncovered the factors affecting classifier performance not known before through use of semantic content clustering and topic modeling.• Wrote data pipelines using Presto Query Engine for processing the data to show on the dashboard. Authored ML pipelines for generating multilingual sentence representations using XLM-R, clustering using K-means and DBSCAN, dimensionality reduction using fast t-SNE and topic modeling with BERTopic.• Trained the multi-modal content classifiers, with modalities of text, image, and video, with a novel weighting scheme based on clusters of content to improve precision (+24% P@R80) and recall (+7% R@P80).

Verif-IDAn ID verification system which reduced 70% manual-work of verifying student ID cards. Trained efficient multi-task CNNs for pre-processing and classifying images into ID cards and non-ID cards. Pre-processing is required for downstream text-spotting models. Read text on the ID cards using DBNet, a text-spotting model, implemented in paddleOCR. Converted text-spotting models from paddleOCR to PyTorch for the task. Tried SDMG-R, a key information extraction model for extracting information from the ID cards and auto-encoders for determining ID similarity. Also tried a few-shot image classification approach as described in the paper FEAT to classify universities of the students. The system is invoked through a CLI application which evaluates a batch of examples with human-in-the-loop for verifying examples.e-Yantra IoT PlatformDeveloped a platform to allow building IoT applications with ease.It has hosted 10+ IoT projects and was used to coach 200+ students and teachers. Was built on top of AWS Cloud using services like EC2, AWS IoT, DynamoDB, and Lambda. Additionally, MySQL was used to store relational data and SNS to send explicitly requested and critical notifications. The front-end was written in VueJS, Blockly (for IFTTT) and Bootstrap.Other projects• Developed a payment intermediary backend in Spring Boot using principles of clean architecture and very high test coverage in unit tests, end-to-end tests and stress tests (written in locust). All the payments happening on several of the edtech platforms of the org go through this system. Implemented a custom spring authentication mechanism based on client secrets and monitoring solutions described by the constraints of the deployment infrastructure.• Built GradeFast++, a platform independent system for creating tests and evaluating programming assignments using FastAPI, Django, PostgreSQL, Redis-Queue, VueJS and Docker. It reduced the in-house evaluators’ work by 60%.

Master'S Degree, Computer Science, 3.8

Coursework: Machine Learning, Computer Vision, Natural Language Processing, Robot Learning, Distributed Systems, Probability and Statistics

Bachelor Of Engineering (B.E.), Computer Engineering, 8.36 Cgpi

Activities and Societies: Member of Robotics and Automation Wing(RAW) at S.F.I.T.

Science, 85%

Ssc, 92.55%