Growth ML Uber TL. Working to deliver on Mark Z's vision: "Over time, I'd like to see us move towards a single, unified recommendation system that powers all of the content including things like People You May Know (friend recommendations) across all of our surfaces." Growth ML at IG is working towards reaching and surpassing the industrial state of the art in content recommendations and then adding aspects that are specific to people relationships. Skills: Leadership · Machine Learning · Recommender Systems

In this role, I worked on 1. unified ranking videos of different video length and format. 2. retention modeling to recommend videos that lead to more successful sessions, users returning for more sessions, more daily active users. We achieved all of the above. 3. debiasing recommendations against being overly affected by popularity of items and from activity of power users. 4. building unified ranking systems that learn from multiple surfaces while retaining the calibration, and uniqueness of each surface 5. multiple representations of the user in retrieval / candidate generation. 6. recommender systems stack of a new immersive video surface Skills: Machine Learning · Recommender Systems Exploring Videos on Facebook Just Got Easier | Meta Video on Facebook Keeps Getting Better | Meta

Support the Homefeed and Notifications ML team. Discord's mission is to create a space that helps people to find belonging online. User experiences we work on this team: - Connecting people to servers and friends they are looking for - Helping people find valuable/delightful moments and content on communities. - Reducing anxiety in delivering the right notifications at the right time to the right user. Levers the team works on: - Server Homefeed - Notifications - Conversation extraction from messages Partner teams: - Growth / Core Experiences - Communities - Apps team - Data Platform - ML Platform - Anti-Abuse ML Using Machine Learning to Build a Delightful Notification Experience Machine Learning at Discord - Gaurav Chakravorty

- Multiple launches in personalized podcast episode recommendations on Google Assistant, ​Google Podcasts app​, Google Discover, ​Assistant Snapshot - - Increased CTR metrics eight-fold, and usage of recommendations from 0 to 30 million monthly active users. - - Quality aspects in recommendations include (a) deep neural networks based retrieval (b) serving-time decision tree and neural networks based reranking - - Worked with Trust and Safety to build a classifier to detect content deemed unsafe for branded recommendation. - - Lead deployment on multiple surfaces, Google Search Universal, Google Discover, Google Assistant, Google Podcasts app - Diversity boosted ranking of recommendations. - Fairness of recommendations - Led development of training data generation pipeline for a personalized neural networks reranking model from user surveys. - ML lead for Endorsements, ‘​Picks for you​’, endorsements that would explain the recommendation and help users reduce selecting poor recommendations. - - Training data generation pipeline. - - Decision tree modeling infrastructure. - - An infrastructure to evaluate multiple models in parallel and hillclimb in quality. - Lead for building a unified modeling framework for end to end personalized recommendations - - Worked on making personalized interaction and logging frameworks privacy compliant and on making new model pushes automated. - Working on an ambitious multi-turn conversational recommendation system. - Led effort to move to Graph Neural Networks for search and recommendations by collaborating with top researchers and presenting to leads. (https://recsysml.substack.com/p/recommendations-using-graph-neural) Nov 2020 to March 2021 Waymo Eval ML team Learnings from UToronto autonomous driving team on an end to end approach of scene embeddings and forecasting Building recommender systems for the creator economy Towards a Graph Neural Networks approach to Recommender Systems

- Built and led teams of about 25 engineers and data-scientists spread across India and US. I am proud of the team I was able to build and how effectively we functioned in a lean budget. - The teams I led: data science, portfolio management, data infrastructure, data management (acquisition and processing and internal APIs), trading infrastructure, research in execution algorithms, technical writing (https://www.qplum.co/investing-library?tab=whitepapers), research, and publications (https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=1427461). - Setup an end to end machine learning based algo-trading system. I architected and participated in the implementation of all parts including data acquisition, data processing, data storage, data infrastructure, research infrastructure, feature engineering, Machine Learning based inference systems, unit and integration tests and robustness enhancements of the production pipeline. - Set up the technology stack [presented here: https://www.qplum.co/stackworld] - Supervise the investments team on live and new investment mandates. - Over 20 Industry presentations and thought leadership events. - Set up a culture of writing code that outlives the tenure of the author. I set up a coding style such that code can be easily understood, linked to company goals via project management tool like Asana, and has sufficient test coverage. - Used: Python, C++, Airflow, Celery, Jupyter Notebooks, SQL, Airflow, Alembic, GOCD, Boost-build, CMake, Docker, Github Pull requests, Asana for project management, lots of whiteboard.

- Architecting a global quantitative electronic trading system. - Setup and management of proprietary trading fund focused on major futures exchanges. - Built a cross-functional team of about 40 data-scientists and software engineers. Each person in the team was found and recruited by me. These are some of the best people in their discipline and some of them followed me into Qplum. Led teams: market data systems, low latency architecture and interprocess communication, machine learning based trading strategy research, data infrastructure, build and testing systems, recruiting and HR. - Used: C++, Perl, Shared memory programming, Python, R, SQL, Airflow, Jupyter Notebooks, Boost-build, Github Pull Requests, Asana for project management, and of course whiteboard!

- I set up a global electronic trading system from scratch. Set up trading systems for over 20 exchanges. I decided the roadmap of the product and worked to meet deployment timelines. - I started with 1 team member and grew the team to about 50 people within 3 years. My leadership style was very hands-on. I was heavily involved in all decisions. Teams led: market data systems, quant research, portfolio construction, risk management, electronic trading systems, business development, external negotiations, recruitment, compensation. - I built an end to end machine learning based system to derive short term price forecasts from data using more than 200 indicators. - The main objective function of the team was to generate profits. As such, I oriented my product management towards that instead of focusing too much on a sub-part. My trading systems made more than $700 MM in profits while I started with just $10K in risk capital in 2005. - Due to my success, I was made a partner in my third year at the firm. I was the youngest ever to make partner. - Tech used: C/C++, Perl, R, BASH Shell Scripts.

Research intern in Computer Aided Verification and Logic Digitizing Interval Duration Logic