Hi, I'm Shreayan. I was born and brought up in Mumbai. I am 27 years old. I am a fun-loving person who likes to travel, eat delicious food, meet new people, and listen to all types of music. I have graduated with a Masters in Computer Science from Johns Hopkins University. I love coding and my areas of interest include Machine Learning, NLP, and Data Science. I started coding in 2012 when my class teacher introduced me to it. My love for coding has only improved over the years. Call me a fanatic, but I'm crazy for all things related to Space and Astronomy. I love researching random stuff in Astronomy and Space, which often gives me existential crisis. My hobbies include playing video games, chess, football and cooking. In my free time, I also play various songs on the keyboard. Do check out my Instagram if you want to check out how I play the keyboard.
Shreayan Chaudhary
+1 (551) 339-7724
shreayan98c@gmail.com
linkedin.com/in/shreayan98c/
Seattle, Washington
I’m a Software Engineer in Machine Learning, specializing in AI agent systems and LLM-powered
applications, with 4+ years of experience building and deploying production-grade AI systems across
industry and research. My work spans agent runtime architecture, distributed multi-agent workflows,
LLM optimization, and scalable ML infrastructure, with a focus on improving reliability,
performance, and cost efficiency.
At Microsoft, I work on the agent harness powering Microsoft 365 Copilot agents, building the
infrastructure that enables reliable tool orchestration, multi-turn execution, safety, and
observability. Previously, at Seagate and Johns Hopkins University’s ARCADE Lab, I developed
enterprise RAG and Text-to-SQL platforms, optimized distributed LLM inference, and built multimodal
AI systems for applications ranging from enterprise data intelligence to surgical robotics.
I’m passionate about bridging the gap between advances in AI research and dependable,
production-ready systems.
Software Engineer II, AI • April 2026 - Present • WA
• Reduced human-in-the-loop (HITL) approval-to-resumption latency by 40%
in Microsoft Copilot through durable checkpointing and concurrent tool approvals across
runtime and backend services
• Instrumented distributed tracing for Microsoft 365 Copilot Agents,
tracking 250K+ service calls across 36K weekly conversations to identify errors and pinpoint
service and network bottlenecks across MCP and A2A interactions
• Strengthened tool-chain security across 58K+ daily calls, reducing
false-positive flags by 64% while preserving human approval controls and data exfiltration
safeguards
Senior Machine Learning Engineer • February 2026 - April 2026 • MN
• Standardized agentic ML infrastructure across 12+ organizations using
Kubernetes and Helm, improving observability, security, and CI/CD, while reducing time to
production by 60%
• Architected a unified enterprise ML platform for RAG and secure
Text-to-SQL across structured and unstructured data, reducing analytics licensing costs by
85% across 4 organizations
Machine Learning Engineer II • June 2024 - February 2026 • MN
• Productionized NL-to-SQL and code-generation agents with sandboxed
Kubernetes execution and PVC-backed isolation, enabling governed enterprise data access
• Optimized Llama 3 time-to-first-token (TTFT) latency to 200 ms at P95
through 4-bit quantization and LoRA/PEFT fine-tuning on NL-to-SQL data, while cutting
inference costs by 67%
Intern Machine Learning Engineer • June 2023 - June 2024 • MN
• Built a multimodal OCR + VLM pipeline with human-in-the-loop validation,
reducing manual purchase order processing by 90% across thousands of POs per week
Machine Learning Engineer • January 2021 - July 2022 • India
• Engineered an OCR and document intelligence pipeline with computer
vision preprocessing, achieving 85% extraction accuracy and improving processing throughput
by 30%
• Developed a code-generation engine that converted user-designed neural
network architectures into PyTorch and TensorFlow code, enabling 40+ enterprise customers to
deploy custom models with automated validation
ML Research Assistant • April 2023 - July 2024 • MD
• Led development of a voice-controlled LLM system for real-time surgical
robotics, optimizing distributed inference (Petals, BLOOMZ, Ray, vLLM) under strict latency
and compute constraints, reducing procedure time by 20%
Machine Learning Intern • December 2019 - January 2021 • India
• Reduced manual translation effort by 80% for Indian Navy by training a
Bi-LSTM Russian-English translation model
• Enhanced resolution across 3.7M+ complaints by applying BERT-based
summarization for NGO feedback workflows
MS in Computer Science • 2022 - 2024
• CGPA: 3.97/4
• Research Assistant at ARCADE (Advanced Robotics and Computationally
AugmenteD Environments) Lab advised by Prof. Unberath - Spring '23, Fall '23,
Spring '24
• Research Assistant at CLSP (Centre for Language and Speech
Processing) Lab advised by Prof. Yarowsky - Fall '22
• Teaching Assistant for the graduate level course Information
Retrieval by Prof. Yarowsky - Spring '24
• Teaching Assistant for the graduate level course Software Engg.
by Prof. Darvish - Spring '23
• Teaching Assistant for the graduate level course Databases by
Prof. Yarowsky - Fall '22
B.Tech in Software Engineering • 2016 - 2020
• Grade: 89%
• First Class with Distinction: Ranked in the top 10% of the
department
XIIth grade in CBSE Board • 2014 - 2016
• Obtained a Silver medal in International Informatics Olympiad.
Xth grade in ICSE Board • 2004 - 2014
• Certificate of excellence in Computer Science and Mathematics.
I started my coding journey in 2012, when my class teacher introduced me to it. My love for coding has only improved over the years. I started coding with Java in 2012. After that, I learned the basics of C and C++. Then I moved on to HTML, CSS and JavaScript to create fluid, mobile responsive and lightweight web applications. When I was introduced to Python by a friend, I immediately fell in love with this language, and it has been my favourite language since. I have dived deep into deep learning (pun intended) and started learning machine learning (also, pun intended) in Python. My area of research lies in Recommender Systems, OCR and NLP.
Shreayan Chaudhary on ResearchGate
Chaudhary, Shreayan Killeen,
Benjamin; Osgood, Greg; Unberath, Mathias (2024). Take a Shot! Natural Language Control of
Robotic X-ray Systems for Image-guided Surgery, International Conference on Information
Processing in Computer-Assisted Interventions
This paper proposes a natural language based communication protocol to control C-Arm robotic devices
using voice commands.
Chaudhary,
Shreayan; Anupama, C. (2020). Ensemble Recommendation System using a hybrid
decision level fusion of Popularity Model and Collaborative Filtering, International
Conference for Artificial Intelligence and Evolutionary Computations in Engineering Systems,
pp.551-559 DOI: 10.1007/978-981-15-0199-9_47.
This paper proposes a hybrid recommendation system algorithm using Content based and Collaborative
Filtering to improve the performance metrics and address the cold-start problem to overcome the
drawbacks of both the algorithms.
Chaudhary,
Shreayan; Ferni, U. (2020). Recommendation System for Establishing New
Businesses using Geospatial Clustering for Multiple Reference Points, National
Conference on Artificial Intelligence and Intelligent Information Processing; Patented under SRM
University.
Created custom clustering algorithms and a recommender system to find the optimal location in any
given city or place to help set up a business for entrepreneurs, thus saving time, money, and risk.
Engineered prototype glasses with camera and mic, and fine-tuned OpenFlamingo on real-time video to
answer users’ audio-based questions (Whisper for ASR) conditioned on the image (CLIP ViT for images
and LLaMA for text)
Data Pipeline: Video Stream from phone camera -> Capture Frames -> Record User Question ->
Transcribe using Whisper -> Model Inference conditioned on image frames -> Return Answer
Technologies: DroidCam, OpenCV, OpenAI Whisper, OpenFlamingo (LLaMA 7B + CLIP ViT/L-14)
Vision-Language Model.
Created an end to end project to help entrepreneurs and businessmen find the optimal location to set
up their business. Created custom clustering algorithms using geospatial data analysis and a
recommender system to find the optimal location in any given city or place to help set up a business
for entrepreneurs, thus saving time, money, and risk.
Data Pipeline: Data Integration -> Data Inspection -> Data Visualizations -> Data Modeling ->
Model Evaluation -> Django Web Application -> Cluster Analysis -> RESTful API creation -> Deployment
on Heroku
Technologies: ML, Custom Clustering Algorithm, Folium, Heroku, Django.
Data Integrated from: Census of India
2011, Geopy Nominatim, Foursquare API and Google Places API.
Venter is a response categorization and document summarization tool. Goal is to create a platform to
analyse, process and summarize large corpora of user feedbacks, inputs and responses
autonomously.
Clients: Civis, ICMyC (I Change My City), SpeakUp.
Technologies: ML, Deep Learning, GCP, NLP, Django.
Created an end to end project to help students find the prospective grad schools for themselves.
Predicting the chance of a student being selected for a grad program abroad and finding the most
suitable universities for him/her.
Data Pipeline: Data Collection -> Data Inspection -> Data Visualizations -> Data Modeling ->
Model Evaluation -> Flask Web Application -> RESTful API creation -> Deployment on Heroku
Technologies: ML, Linear Regression, XGBoost, Heroku, Flask, AWS.
Created an automated machine learning model that will predict the propensity of a loan defaulter
whether he will return the loan (s)he has taken from the bank or not and find out the most probable
form of repayment. Also automated it to run at 6pm daily and deployed it on an AWS EC2 server for
production.
Creating a smart headband which is connected to the headphones to guide the visually impaired user
via audio instructions about the objects and people in his/her path. This project was awarded a
special mention at Conception, an IoT project expo.
Created a recommender system by applying collaborative filtering and simple popularity method
algorithms on a movie dataset consisting of 27000 movies and 20 million ratings given by 138000
users.
Dataset: MovieLens 20 Million
Dataset
Created a Smart Vehicle to Vehicle Technology using IoT to decrease commute times as well as enable
multi-vehicle control by a single driver using GPS and Ultrasonic sensors.
Technologies Used: Keras, Arduino, NodeMCU, Bluetooth, Flask
Languages Used: Python, HTML, JavaScript, CSS, C
Creating a local library application using django to manage books systematically. This project was
created in Django and was deployed on Heroku server. Project on GitHub
Used the techniques of SMOTE (Synthetic Minority Oversampling TEchnique), under-sampling and
over-sampling to classify the type of business in multiple categories in an unbalanced dataset and
achieved top 10 percentile F-1 score across India.
A news blog using django that fetches the latest news via news blog api. This project was created in
Django and was deployed on Heroku server. Project
on GitHub
Created a deep neural network using keras to identify the type of cloth or the footwear from the
given images.
Using the OpenCV library to detect the user's face from the frames of the video captured through the
videocam.
A web application which can send Emails, SMS and Android push notifications to application users to
all (or selected few) employees in an organization. It uses HTML, CSS and JS as the frontend for
webpage and MSSQL Server and Java as backend which stores the employee details.
Software Requirements: Netbeans IDE 8.2, Java 8 EE, Microsoft SQL Server 2012
Languages Used: HTML, JavaScript, CSS, Java
Frameworks Used: Bootstrap, DataTables, jQuery
To predict the price of a house in Boston using scikit-learn’s boston dataset.
The problem that we solved here is that given a set of features that describe a house in Boston, our
machine learning model must
predict the house price.
Languages Used: Python
Dataset: Boston
Houses Dataset
Created a model using machine learning algorithms that can predict whether the cancer is benign or
malignant.
Dataset: Breast
Cancer Wisconsin Dataset
Objective was to understand the factors that influence the attrition and to predict the employees
who are going to leave the company in future.
Dataset: IBM Employees
Dataset
A very simple to-do list that allows you to keep a track of activities / tasks you need to complete
without having to forget about it.
Check it out!