Sidharth Gupta Email & Phone Number

New York, Ny, Us

New York, Ny, Us

Finsbury Park, England, Gb

Deep learning: My favorite project was augmenting Dan Cruickshank's wonderful London to help me discover unique experiences that I threaded(see below)A very finetuned-RAG take on Cruickshank's work.More formally it was a wonderful peer led onsite bootcamp for deep learning program supported by the UK Government in collaboration with Founders and Coders (https://ml.institute/) for building of AI talent at the national level.I learned everything starting from Word2Vec to Sequential Networks to Real Time Object Detection(https://arxiv.org/abs/1506.02640) to Coherent and Minimal story telling models(https://arxiv.org/abs/2305.07759). The crucial part was reading the papers and implementing them in pytorch or huggingface and learning how they worked.

We all travel differently and plan, select, execute plans very differently from our previous generation. Real life travel has to be customized to our preferences and values are the can't just be cookie cutter any more. Finding the right market fit and scaling a product like this is not easy and that is what we want to achieve. - Established a system for Retrieval & Ranking pipelines using RAG like flows to support In Real Life(IRL) agents. Agents are your BFFs and they use your Context, Preferences, and past Events' interaction history to help you choose fun things to do.- Advised the team on how to plan and execute and put together a PoT for validating product use cases.

Precise and Coarse Location inference is finding the intersection of a place and a person's signals. At meta I explored various facets of its application: From Lift Measurement(B2B) to Wearables(B2C) products.- Feature engineering, error Funnel Analysis, Sensor(BLE/Wifi) deployment Opportunity analysis for Place inference Model Improvement, upstream to mission critical Store Visit Ads Lift measurement. - Understand work and Baseline Multilabel models using Pytorch Activity Recognition for FaceBook Reality Labs. - Ground Truth data collection to support model development for wearable products for FaceBook Reality Labs supporting products leveraging Location and Place inference data for Fitness and Place Reminders.- Led several privacy initiatives to safeguard sensitive location data for FB users.

New York, Ny, Us

Completed 2 courses in the Columbia Statistics Department:Bayesian Statistics for Social SciencesLinear Regression ModelsUnderstanding of parametric inference techniques.Understanding of Regression modelling and Hierarchical Regression.Learnt Bayesian modelling and diagnostics using Stan and R. Developed understanding of inference and model selection using Full Posterior Predictive distributions.

I worked with a good friend to help TR optimize Development flows for Search Applications in TR. A bit too opsy for me, but I made good friends.- Optimized Development Process: A reduction in development & testing time, by refactoring the team's build & ship process to include an Alpha environment by designing golden dataset & using Gradle, GitLab for seamless build experience in Alpha.- Enhanced Data Management & Testing: Improved data linkage & testing efficiency through the implementation of automated data triggers in the alpha environment, ensuring seamless integration of code & data updates.

New York, Ny, Us

Accelerated Image Search Availability: Increased Shutterstock's image Solr index refresh rate, enabling faster search & purchase of new inventory, by developing a multithreaded Java application for efficient data indexing from key-value store to Solr.Successfully optimized Solr indexing & search performance, resulting in scalable & more efficient data handling, by tuning the Solr JVM with the G1 garbage collector.Enhanced the performance & accuracy of new ranking functions, as evidenced by improved API response times & relevancy metrics, through performance testing with JMeter & log replay.

Hoboken, New Jersey, Us

Productionized the Student Engagement model for Knewton's Adaptive Recommendation platform by implementing the Gaussian Mixture Models for Knewton's Adaptive Recommendation Engine, this was achieved through implementation in EMR.- Learn't the fundamentals of Gaussian Processes to model temporal student engagement and alternatively using Gaussian Mixture Models for log normal response times.- I also implemented scaled implementations of these models to learn offline parameters on large datasets using Hadoop (MRJob) and Apache Spark.In house Zookeeper based Service Monitoring and Discovery in JAVA for distributed Service Discovery and Config Management.- Designed curator client wrappers with caching to improve on zookeeper availability.- Service/Config discovery and Publishing implemented to support SOA at Knewton.Readiness Platform Dashboard: Created a light web application backed by existing RDS views to display insights into cohort performance and content analytics on our Math Readiness platform. The feedback was crucial to improve Knewton's Readiness Platform's instructional and assessment content.Analytics for model performance feedback to data scientists (2014): Led the development of analytics tools for monitoring machine learning models in production, improving understanding of variance changes & key contributing factors for model degradation, through in-house analytics development in Looker.

Parse.ly(Technology Startup), NY:*Objective1: Build from scratch a personalized News Reader Application to recommend online print media content to match a reader's interests. - Frontend Development: Used Django with jQuery, CSS Grids(960.gs). - Backend Core: Mapped user interest and web content to Wikipedia ontologies, stored in mongodb. This enabled fast recommendations for users by inference on ontologies. User data gathered from Twitter and Facebook via Oauth.*Objective2: Consolidation of internal analytics metrics and elimination of redundancy of named entities in databases.- Leveraged Wikipedia and Dbpedia datasets to perform Named Entity Normalization, alias detection, Word Sense Disambiguation. Preprocessed data using POS tagging, fuzzy string matching with edit-distance.Public Clearing House(PCH) systems, NY:*ExtJS refactoring: Refactored sizable ExtJS code base (50+ JS classes):- I applied design patterns to reusable components like data stores, tree panels, grids , toolbars/action menus etc. This increased usability and reduced the code base size by 30%. I learned ExtJS from the ground up in 10 days.- Built new screens and functionality in ExtJS, working from wireframe specifications.

Tempe, Az, Us

Worked with Dr. Seungchan Kim on Gene Regulatory Networks for improving classification models for small sample and noisy data. Masters thesis work was the development of a variant of maximum margin classifier that is suited to address the lack of robustness of discriminant based classifiers to noise and outliers.

*CORE Banking project (September 2005 – May 2006)- Load balancing of legacy applications in HP-UNIX and development of multi threaded non blocking server in C++ for high throughput network loads. Hands on experience on multi threading in C++ using the p-threads library, STL classes, OOP & design patterns.*Web Portal maintenance for General Electric Corporate (May 2006 – May 2008)- Web applications optimization techniques by developing mini AJAX based framework, basic data-proxy and custom XML protocol and a REST web-service. - Separation of business and presentation logic on critical web pages to reduce processing redundancy of HTTP requests. Reduction in the latency of web service to enhance user interaction. Interfacing application with j2EE/PHP frameworks.

M.S, Computer Science

B.E, Electronics And Instrumentation Control

Education record