FlowCV Logo
Abhijith CLead AI Engineer
Professional Experience

Lead AI Engineer, Data Sciences

05/2024 – presentBangalore

Staff Machine Learning Engineer 1

Glance, InMobi
  • Led the design and development of a real-time recommendation and content serving system using (PySpark) Structured Streaming, and gRPC protobuf for communication over Kafka. [blog]
  • 01/2024 – 05/2024Bangalore
  • Led and managed a high-performing team in the end-to-end development of recommendation systems, serving a user base of 100+ million daily active users while maintaining sub-500ms latency.
  • Applied RL techniques such as MABs to optimize user experiences for less-engaged and sporadically active users using Thompson Sampling, beating our control by ~20%.
  • Served as the key point of contact for a cost-saving initiative to minimize cloud expenses for various machine learning and data engineering workloads, saving upto 50k USD per month.
  • Led the design and successful execution of PoCs focused on enabling machine learning on mobile Edge devices using TFLite.
  • Collaborated with MLOps and Engg. teams to implement production-grade practices for ML systems, with monitoring, logging, and performance testing, primarily on GCP. [blog]
  • Designed, built and iterated on Candidate Generation layers for an 'online-dynamic-feed' system using ANN methods.
  • Machine Learning Engineer 1, 2, 3

    Glance, InMobi
  • Led the development of an automated content creation platform, generating over 95% of the content published on Glance, using a Engg. and ML stack built from scratch.
  • 08/2019 – 12/2023Bangalore
  • Developed multiple image and text based automated solutions for deduplication, scoring, image ranking and search, image modifications/edits and publishing for the content creation platform.
  • Led the efforts on building a content ranking engine that scores potential content for enabling strategic content selection for Glance.
  • Deployed real-time streaming systems, leveraging Azure Eventhub and GCP pubsub, to enable content annotations and data enrichment of 'Glances' which helped several teams optimize their processes. Still the backbone of our Rec Sys.
  • Developed a system to automatically convert landscape videos into portrait videos without losing out on the 'busy' or important scenes/objects of the video (based on some Google Apis, before AutoFlip was released).
  • Developed and deployed Audio Classification and Content Moderation models for Roposo, a short form video application.
  • Data Scientist

  • Addressed image blurriness issues in ad placements by developing predictive models for optimal image-to-slot sizing, alongside the implementation of super-resolution models.
  • 2018 – 2019Bangalore
  • Led the creation of 'Creative Insights', a comprehensive solution that involved the evaluation and ranking of ad images. This was complemented by an Insights Engine designed to substantiate recommendations for enhancing ad image quality, used combination of CNNs and hand crafted features for button detection.
  • Increased the operational efficiency of Ad moderation by introducing Ad de-deduplication (SIFT, Siamese Nets) and bad-ad detection computer vision systems (CNNs)
  • Education

    Bachelor of Engineering, CS

    Sir M. Visvesvaraya Institute of Technology
    2014 – 2018Bangalore

    Master of Science, CS (ML Specialisation)

    Georgia Institute of Technology
    2022 – 2024

    Mastermind in Data Science


    Multiple Quaterly Best Performer Awards

    InMobi + Glance

    Honorable mention, ACM ICPC 2016 & 2017


    Google Summer of Code, with Python Software Foundation


    Biometric Signal Sensor’s Interface for Astronauts

    Google Summer of Code Project

    GSoC 2017 project under Italian Mars Society (sub org. Under Python Software Foundation)

  • This project mainly revolves around configuring biometric signal sensors
  • Using Hexoskin smart shirt as the device used to monitor astronauts
  • Integrating the the smart shirt (the sensors) with Tango-Controls Server
  • SQLAlchemy ORM for the database
  • Image Regeneration with Generative Models

    Final College Project
  • An approach to use CapsNet as a discriminator in Generative Adversarial Network and demonstrate its appli- cation using semantic inpainting on MNIST and face images.
  • Minesweeper (C++)

    College Project
  • This project demonstrates a simple game of Minesweeper.
  • Built using OpenGL Graphics Library (freeGLUT) in C++
  • GoChat

    College Hackathon Project
  • An universal linux application that let’s it’s users to chat with one another through the terminal
  • Packaged using Snapcraft and hosted on AWS ( Serverless - AWS Lambda ), for the communication
  • Several Android applications, libraries

    College Hackathon Projects
  • Textify - An android application that extracts text from images and screenshots, used FreeSpaceOCR API
  • Listview-Snapshot (Android Library) - Android library that converts listViews to Snapshot images
  • AuReader - Android Ebook reader that plays music based on the emotion of the context, used Natural Language Processing(NLP) library, synesketch
  • WhatsCrap - An application that lets you delete old WhatsApp backup databases so that your phone storage space is saved