Mar 2023 – Apr 2024 · 1 yr 1 mo · London Area, United Kingdom

• I own the Content Journey charter where I managed xx members comprising of Managers, MLEs and SDEs spanning across 3 teams. The Content Journey charter powers recommendations(Text based, image, video etc) across 16 languages and xx Million DAU. Primary responsibilities involved driving user retention and improving user engagement on ShareChat platform (valued at $5B).
• A few initiatives that I owned/started:
• 1. Online Learning algorithms which learnt user and item embeddings from 100k+ Requests per second. Challenges involved tackling RecSys goals (popularity bias, diversity, relevance, Recency aware recommendations etc) and Cost goals (stream processing costs, storage costs, compute costs etc).
• 2. Feature rich embeddings: Learning embeddings which incorporated user and item side information. Introduced Two Tower models to our stack.
• 3. Recency aware recommendations: Introduced algorithms which incorporated user's in-session information to quickly adapt. Involved working on real time online learning algorithms, incorporating in-session negatives with triplet loss etc.
• 4. User x Post A/B framework for supply side experimentation: User A/B is typical A/B testing framework for evaluating personalization experiments. However, they cannot evaluate supply side interventions which affect post progression metrics (0 to 60k + views), creator interventions (boosting) and genre/tag diversity interventions. Involved in proposal, architecting and building this framework from grounds-up and proposing metrics to track aforementioned interventions.
• 5. Faster offline evaluation: Offline evaluation of Candidate Generators is hard as heavy ranker acts as the biggest confounder in getting true labels. Led the development of multiple metrics and a framework for tackling this issue.

Feb 2022 – Feb 2023 · 1 yr · London, England, United Kingdom

• Tech Lead for the Representations Learning team which powered Recommendations across Home TimeLine Ranking and Magic Recs aka Twitter Notifications. Responsibilities involved building ML models to achieve business targets (UAM increase, mDAU increase), frame technical roadmaps to better the RecSys charter and mentoring junior ML talent.
• Spearheaded the following intiatives:
• 1. Improving recommendation quality of Light Users: Light Users typically don't yield explicit signals during their session like clicks, likes, shares etc, thus making it difficult to personalize their feed. Built follow based embeddings to elucidate user interest segments and thus yielded better personalization.
• 2. Out of Language recommendations: Preventing recommendations occuring in languages that user was not used to.
• 3. Out of network recommendations: This involved tackling diversity goals wherein users were recommended posts from creators typically not in their follow networks.

Sep 2019 – Dec 2021 · 2 yrs 3 mos

• Responsible for driving exploration on in-session user signals for personalizing Booking.com's core Ranking algorithm. Recommendations at Booking.com typically suffer from continuous cold start (user behavior changes every holiday season, users don't sign in etc). The in-session signals were used to learn better user and hotel embeddings to improve the recommendations. Responsibilities involved mentoring a geographically distributed team to achieve Conversion goals.

Nov 2017 – Sep 2019 · 1 yr 10 mos

• Founded the team working on detecting bots in our traffic(Booking.com is one of the largest e-commerce giants on the planet). Bots aka automated traffic raise costs(Pay Per Click Ads, infra costs) and wreak havoc on internal metrics/insights, experimentation, data quality for training ML models etc. Owned and lead the charter for building bot detection models from grounds up and managed to have an impact. Gave a bunch of (plenary) talks on this as well.

Apr 2017 – Oct 2017 · 6 mos · Bengaluru, Karnataka, India

• Focused on ranking and dynamic pricing problems for hotel booking

Mar 2015 – Mar 2017 · 2 yrs · Bengaluru Area, India

• I was part of the Click Prediction team at Microsoft Bing Ads. Worked on the problem of Click prediction for paid/sponsored search for EU market. Core responsibilities included building models for predicting Probability of Click, Probability of GoodClick, Dynamic bid modification etc. Got to work on Petabyte scale datasets which challenged conventional forms of ML.

Jul 2014 – Feb 2015 · 7 mos · Bangalore

• Part of an internal startup called CeTI within the CTO's office. Led a team of 6+ people (FTEs+contractors). A few notable projects include:
• 1. EMC LifeCare: Predictive analytics platform for patient centric healthcare
• I was responsible for the core engineering as well as building the Machine learning/Analytics applications for this platform. Prototype was tested in two clinics of the Apollo chain in Hyderabad and Bangalore.

Jul 2012 – Jul 2014 · 2 yrs · Bangalore

• Part of CTO Office's CeTI initiative which is an innovation hub within EMC tasked with building Go-To Market products.
• A few notable current projects include:
• 1. Building a planner for automating EMC's data center migration business: Single-handedly built the planner. In recognition of my efforts, was awarded the Bronze medal for excellence at EMC. The planner was productized by EMC and a patent on the underlying algorithm has been granted by USPTO. I was awarded Equity grants for this contribution by the CTO of EMC India COE