
Building the future of AI Safety | Ex - Meta, Roblox
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
University at Buffalo
Master of Science (M.S.), Computer Science
January 1, 2015 – June 1, 2017
University of Mumbai
Bachelor of Engineering (B.E.), Electronics Engineering
January 1, 2010 – January 1, 2014
Wrynx
Founder
February 1, 2025 – Present
Roblox
Software Engineer
June 1, 2023 – September 1, 2024
San Mateo, California, United States
Meta
Software Engineer
April 1, 2020 – March 1, 2023
Menlo Park, California, United States
Pivotus
Machine Learning Engineer (Research)
October 1, 2017 – April 1, 2020
San Francisco Bay Area
Pivotus
Machine Learning Intern (Research)
June 1, 2017 – August 1, 2017
Menlo Park, California
ACV Auctions
Data Science Intern
December 1, 2016 – May 1, 2017
Buffalo/Niagara, New York Area
Sungevity
Software Intern (Site Reliability / DevOps)
May 1, 2016 – August 1, 2016
Oakland, CA
University at Buffalo (UB)
Student Research Assistant
November 1, 2015 – January 1, 2016
Buffalo/Niagara, New York Area
Handwriting Matching Using Deep Learning
September 1, 2016 – October 1, 2016
• Achieved 91% accuracy in matching two people's handwriting by implementing a convolutional neural network based model. (Python/TensorFlow) • Achieved a maximum of 73% accuracy by implementing an auto-encoder + fully-connected neural network model.
Creating reusable holdout sets using Differentially Private Algortihms
April 1, 2016 – May 1, 2016
Successfully generated multiple holdout (validation) data sets from a single holdout (validation) sets by implementing the differentially private thresholdout algorithm.
MAC Random transmission Protocol Simulation using NS-2
December 1, 2015 – December 1, 2015
• Achieved 17% successful transmission rate for my protocol implementation in a simulation environment containing 100 source nodes and one sink node. • Developed a script in tcl to give the protocol specifications. • Implemented the MAC Random transmission protocol in C to work in conjunction with the NS-2 simulator. • Developed a script in Python to analyse the trace results and calculate parameter values.
Handwritten Digit Recognition
November 1, 2015 – December 1, 2015
• Developed a solution to identify digits from handwritten image samples using: a)classical regression model b) Artificial neural network and c)Convolutional neural network(Deep Learning) • Achieved a classification rate of 96.4% on unseen data using the classical regression model. • Achieved a classification rate of 98.2% on unseen data using the artificial and Deep neural networks. • Modeled the structure, hyperparameters and link weights of the neural network using MATLAB. • Trained the classical regression model and tuned hyper parameters using mini-batch stochastic gradient descent method to determine the appropriate weights and normalization factor.
Eye-Blink driven keyboard
November 1, 2015 – November 1, 2015
Won the M&T Bank award for 'Best Hack That Improves Accessibility to Financial Services for People Living with Disabilities' at UB Hacking 2015. • Conceptualized, designed and implemented a system to capture user input through eye blinks • Successfully implemented C (Open CV) code to a) Detect eye blinks b) Create a FIFO pipe and c) Write data to a FIFO pipe. • Implemented a program in Java to read data from the FIFO pipe in real time and then simulate keyboard key presses.
Multilingual Tweet Search Engine
November 1, 2015 – December 1, 2015
• Designed and implemented a full-stack solution to search a given corpus of multi-linugal Tweets (English, German, French and Russian) through cross-linugal queries. • Successfully implemented facted search, content tagging and Latent Dirichlet Allocation (LDA) topic modeling. • Performed language detection and translation both at run time and index time using Java, JQuery and Lucene. • Scraped Twitter using the REST API in python to collect tweets in JSON format. • Performed content tagging and LDA topic Modeling on the tweet data using Alchemy API in Java. • Developed a front-end interface for the system using AjaxSolr (jQuery), HTML and CSS. • Hosted the entire website on the cloud using Microsoft Azure
Ranking System using Linear Regression
October 1, 2015 – October 1, 2015
• Implemented a system to rank entities using linear regression models using Gaussian radial basis functions. • Tuned the hyper parameters and weights using closed form and stochastic gradient descent methods on Microsoft LETOR (LEarning TO Rank) dataset.
Search System using the Boolean Model
October 1, 2015 – Present
• Indexed approximately 25000 term-Postings pairs in memory using Java • Developed a method to perform Term-At-A-Time AND/OR Boolean queries on the data. • Developed a method to perform Document-At-A-Time AND/OR Boolean queries on the data.
Routing System Implementation
October 1, 2015 – November 1, 2015
• Developed a system using C(gcc) to simulate the working of Distance Vector Protocol • Programmed each node on the network to function as a router. • Programmed UDP sockets to facilitate communication between the different nodes in the network • Developed a user interface to enable users to manually access parameters and control the functioning of the protocol • Developed a method to construct the network topology from data stored in a local file.
Twitter Crawler and Indexer
September 1, 2015 – September 1, 2015
• Developed an application in Python to get and parse multi-ligual tweets (English, Russian and German) tweets via the Twitter REST API. • Parsed the JSON encoded data to extract relevant information • Indexed the JSON-encoded data using Apache Solr on Linux. • Implemented generic and language-specific tokenizers, analysers and filters to optimize search results by handling compund words in German and transliterated queries in Russian.
Distributed File Sharing and Chat System
September 1, 2015 – October 1, 2015
• Developed a system to share files over a client/server and peer-to-peer hybrid network. • Programmed TCP sockets using C to communicate between nodes. • Used the select() API to handle multiple socket connections and console input on a single thread.
University Data Analytics using Bayesian Networks
September 1, 2015 – September 1, 2015
• Designed a graph search based algorithm to determine the best relationship between a given set of parameters using Bayesian Networks. • Implemented the above algorithm using MATLAB to read data values from an Excel spreadsheet, compute means, variances, standard deviations, log likelihood and programmatically construct the most optimum Bayesian network to illustrate the relationship between the variables.
Contactless Credit Card payment system
August 1, 2013 – April 1, 2014
Built a system to process card transactions at retail outlets without physically swiping the card using 8-bit ATMEGA168 microcontroller (master), a wireless LAN system in package chip (HDG104) and an Android app. ▪ Programmed the app in Java to to scan a QR code, extract billing details, accept/store user card details, encrypt them using and transmit them back to the hardware device over WLAN using 802.11g and WPA2 personal encryption. ▪ Programmed the microcontroller using Java (Arduino)
SMS Studio
December 1, 2012 – December 1, 2012
▪ Developed a desktop application (Windows/Linux) in FreeBasic to interface a mobile phone/GSM modem through USB and send bulk SMSes using AT commands(Hayes Command Set).
Standalone LED Animation Display Board
January 1, 2012 – February 1, 2012
▪ Designed and implemented algorithms using Java on Arduino to display moving animations on a self-built LED matrix display. ▪ Designed a LED matrix display by cascading lower order demultiplexers(decoders) and using current sinking devices.
Standalone Digital Scoreboard
January 1, 2012 – February 1, 2012
▪ Designed an algorithm and implemented it on an embedded device using Java to parse ASCII data read via USB-UART and display it on a custom made LED display board. ▪ Designed a LED display board (6 panels consisting of 6 7-segment displays each) by cascading lower order demultiplexers(decoders).
Call Recorder
January 1, 2012 – January 1, 2012
Won 2nd prize at ASME SDE, a hardware design competition conducted by IIT Kharagpur ▪ Designed a call recording solution using MATLAB and a few passive electronic components for less than INR10 ($0.15)
Cultural Fit Analysis
The candidate demonstrates a strong cultural fit for an ML Engineer role, particularly in environments focused on innovation and complex problem-solving. Their diverse project portfolio, ranging from academic research to practical system implementations, showcases adaptability and a broad technical curiosity. Experience in leading teams and mentoring interviewers at large companies suggests a collaborative and growth-oriented mindset. The focus on 'Making AI models safe' at Wrynx aligns with ethical considerations in AI development.
Soft Skills & Operational Fit
The candidate's experience at Meta and Roblox, including leading migrations and significant involvement in recruiting, suggests strong leadership, collaboration, and communication skills. Their role as a founder at Wrynx also indicates entrepreneurial drive and problem-solving abilities. The project descriptions are clear and detailed, indicating good written communication.