Yogendra Yatnalkar 💡 #


Connect Over LinkedIn yogenyat@gmail.com
Follow Over GitHub YouTube

About Me: #

I work as an AI Engineer @ Google (GCP) with four years of industry experience. Previously, I worked at Quantiphi Analytics as a Senior ML Engineer. I have in-depth experience with AWS and GCP, as well as some experience with Azure.

Since I recently started at Google, I’ll update the information below in the coming months. For now, here is an overview of the work I did at Quantiphi:

  • ML Engineer: Developed custom ML/DL solutions for global clients, with hands-on experience in:
    • Tabular Data: Churn prediction, classification, and other similar use cases
    • Computer Vision: Expertise in segmentation, object detection, image similarity, video deduplication, video analytics, and more
    • MLOps: Extensive work with SageMaker Jobs, SageMaker Pipelines, and Airflow, including building components for processing, training, monitoring, and explainability
    • NLP: Built LLM- and RAG-based chatbots using AWS Bedrock and AWS OpenSearch
  • Research Engineer: Primarily focused on Computer Vision research with two main activities:
    • Exploring new techniques and papers to identify potential applications within ongoing projects
    • Evaluating the performance and usability of new AWS ML services
  • Pre-sales ML Consultant: Supported the sales team by engaging with customers, understanding their challenges, and proposing solutions, particularly for Computer Vision use cases.

Key Technical Achievements #

  • Developed and deployed a custom GAN model for shadow generation below objects in early 2022
  • Acted as one of the core developers for Quantiphi’s MLOps product on AWS: NeuralOps
  • Research conducted on:
    • Backtracking in AWS Lookout for Vision (recognized by AWS Community Builders for this)
    • Multi-Task Learning
    • Prompt-less finetuning of MetaAI’s Segment Anything Model (SAM)
    • Balancing loss impact during ML training when using multiple loss metrics
    • …and more
  • GPU Experience:
    • Trained a salient image segmentation model for seven days on a single GPU in 2021
    • Explored multi-GPU training with Horovod, PyTorch DP, and PyTorch DDP
    • Hosted large ML models on AWS Inferentia
    • Used FFMPEG on GPU for large-scale audio-video transcoding
    • …and more
  • Working at Scale:
    • Worked on 1.5 TB of CSV data for a churn prediction use case
    • Managed 70+ TB of raw audio-video data for a content deduplication use case

More About Me: #

Hey there, if you have reached till the end then you can visit the following pages: