Tanya Chutani — AI Researcher

"Hello World"- We have adopted this phrase as a sign to greet, sustenance, creativity and a sign of life. This serves as an apt expression to breathe life into the following sentences as I begin to describe my journey as a Data Scientist. Through a mixture of Coursera and self-learning, it was during my bachelor's that I recognized that the ability to process and interpret data in an efficient manner presented us to make the world a better place by providing us with the power to make data-driven decisions. Following some setbacks in my journey early on in my quest, I kept on laboriously researching the latest trends and solutions to stay abreast, recognizing overnight success stories are few and far between. Somehow, through a combination of self-interest and acquainting myself with great minds, I came across the fascinating world of Computer Vision which enables machines to replicate the human visual system, signalling a feeling of "Eureka!" in me- a reason for my being and a never-ending passion to yearn to learn, exploring and tackling problems in this domain. I have extensively worked on implementing cutting-edge technology directly from research papers, and strive to be a continuous practitioner of learning by doing philosophy, thus aiming to provide value to any project I am tasked (or task myself) upon. It also entails that in my spare time, I love to read research papers and endeavour to code simplified implementations of them using the Tensorflow framework for deep learning. Having said so, I have been quite the fortunate one, having had a management structure and teams around me to assist me, with the best exhibit being my current set-up here at Eagleview. I am responsible for developing applications centred on computer vision to refine and create solutions for multi-scale geospatial object recognition and segmentation in high spatial resolution remote sensing satellite imagery. My previous work experience in this domain includes having worked in the medical domain at Sigtuple, enhancing catalogues and recommendations of fashion stories at Charmboard, and developing end-to-end forecasting pipelines at an agrotech startup, Credible India. Open source platforms are vital for technology agility, facilitating the free exchange of ideas within the developer community. I love gaining and sharing my experience on GitHub and contributing on a regular basis. I have created applications using Vision Transformers, Model Quantization, Self-supervised learning and GANs. You can connect with me through email at - tanyach1997@gmail.com.

Stackforce AI infers this person is a Data Scientist specializing in Computer Vision and Machine Learning across Healthcare and Geospatial industries.

Location: Mumbai, Maharashtra, India

Experience: 6 yrs 3 mos

Skills

Mlops
Deep Learning
Computer Vision
Machine Learning

Career Highlights

Expert in Computer Vision and Deep Learning.
Proven track record in developing scalable AI solutions.
Strong background in MLOps and automation.

Work Experience

AIRA Matrix Private Limited

Lead AI Engineer (2 yrs 9 mos)

EagleView

Data Scientist II (2 yrs 2 mos)

SigTuple

Data Scientist I (6 mos)

Charmboard

Data Scientist (9 mos)

Ocwen Financial Corporation - US

Data Analyst Intern (1 yr)

Credible India

Machine Learning Engineer Intern (2 mos)

Education

Master's degree at Birla Institute of Technology and Science, Pilani

B.E. at Chitkara University

12 at New Yashoda Public School

Tanya Chutani

AI Researcher

Mumbai, Maharashtra, India6 yrs 3 mos experience

Most Likely To SwitchAI ML Practitioner

Key Highlights

Expert in Computer Vision and Deep Learning.
Proven track record in developing scalable AI solutions.
Strong background in MLOps and automation.

Stackforce AI infers this person is a Data Scientist specializing in Computer Vision and Machine Learning across Healthcare and Geospatial industries.

Contact

tanyach1997@gmail.com LinkedIn

Skills

Core Skills

MlopsDeep LearningComputer VisionMachine Learning

Other Skills

AlgorithmsAnalyticsBERTCC++Convolutional Neural Networks (CNN)Data AnalysisData AnalyticsData ScienceData VisualizationForecastingHTMLImage ProcessingJavaScriptKeras

About

Experience

Aira matrix private limited

Lead AI Engineer

Jun 2023 – Present · 2 yrs 9 mos · Mumbai, Maharashtra, India · On-site

MLOpsDeep LearningComputer Vision

Eagleview

Data Scientist II

Apr 2021 – Jun 2023 · 2 yrs 2 mos · Bengaluru, Karnataka, India

Identification of multiple structures
Built an end-to-end pipeline for segmentation of multiple structures and
modified model architecture and loss function in order to make it compatible
with sub-classes provided in the dataset.
Performed model int8 quantization to reduce the inference time and mixed
precision training was added to reduce training time.
Inculcated best MLOps practices to lower technical debt and adapt workflow
that requires little to no manual intervention for deployment and monitoring.
Poly loss and test time augmentation helped in increasing the metrics by 1/4th.
Post-processing techniques (morphological operations) were applied to remove
small objects from the predicted model. Lastly, served the deployed TensortRT
model on AWS sagemaker.

MLOpsDeep LearningComputer Vision

Sigtuple

Data Scientist I

Aug 2020 – Feb 2021 · 6 mos · Bengaluru, Karnataka, India

Urine Sediment Detection
Built an object detection model for detecting 7 categories of urine sediments and
deployed on the AI-100 device via docker container for automatic microscopic
urinalysis to increase the throughput of doctors and reduce the need for intensive
manual examination.
Trained a custom YOLOv5s model on patches of the microscopic urine images
with a removed background which led to an increase in mAP by 7%.
QI Model
Developed a model capable of morphological analysis of urine sediment to filter
out the insignificant low-quality medical images with amorphous materials used
for quality improvement of the models for detection of RBCs, WBCs, etc.
Implemented Knowledge Distillation model compression method to transfer
knowledge from ResNet as a teacher to MobileNet as a student model which
led to a decrease in inference time from 4 to 2.5 minutes which facilitated the
deployment of the model on portable devices at hospitals.
Used GradCAM technique to create visual explanations of the regions in the
input microscopic images which were deemed important for predictions that
helped to improve the trust of diagnosis with doctors.

Charmboard

Data Scientist

Nov 2019 – Aug 2020 · 9 mos · Bengaluru, Karnataka, India

Automated tagging of Fashion Images
Built a tag generation system using a combination of images and text from
fashion cards and charms which were able to replace an existing rule-based system
and eliminate manual tagging.
Trained Xception over fashion images from 127 tags, fine-tuned BERT model on
top of charm captions and subheadings which were used to predict tags for a
particular fashion card.
Developed pipelines to improve workflow by efficiently handling data and
optimize model training phase using tf.data and accelerated linear algebra (XLA)
with mixed-precision training respectively, which reduced the total training time
by a factor of one-third.
Auto Board Categorization
Implementation of a fast and scalable fashion catalog system from board
descriptions and meta tags with the aim to lessen the user click to search for an
apparel.
Used bidirectional GRU on top of embeddings generated from word2vec to
classify fashion boards into 20 board categories.
This approach to categorize fashion boards led to an increase in content
discoveries by 24% user engagement by 8.6 seconds for more than 1 million
users.

Ocwen financial corporation - us

Data Analyst Intern

Sep 2018 – Sep 2019 · 1 yr · Bengaluru, Karnataka, India

Automation and enhancement of reporting for long-established processes for the foreclosure business units which led to more time being devoted to sporadic requests.
Coordinated reporting with the business units to ensure business continuity and smooth transitioning during the PHH-Ocwen merger to successfully integrate data for millions of loans into existing databases.
Created routine reports and documented compliance changes in coordination with foreclosure business units that benefited daily production activities and improved reporting of operational processes.

Credible india

Machine Learning Engineer Intern

Jun 2018 – Aug 2018 · 2 mos · Remote

Responsible for development of a risk management tool for smallholder entrepreneurs in a rural cooperative to empower agribusinesses using forecasting techniques that helps them scale-up with minimum resources.
Developed a forecasting model using Stacked LSTM sequence-to-sequence autoencoder to predict weekly prices of chicken which helped drive sale strategies for women poultry farm holders in Maharashtra, under supervision of co-founder Pooja Rao.