Apr 2022 – Sep 2024 · 2 yrs 5 mos

• 1. Managed team of 12 members across 3 major projects for ML team in Freshworks CRM product. Role included architecting ML systems, model building and experimentation. Having regular career discussion with reports, managing team budget, ensuring timely completion of projects. Apart from that I use to spend 20% of time working hands on.
• 2. Zero attrition in team in 2022. Received 95%+ approval rate from team members in team surveys.
• 3. We were granted one patent ( https://tinyurl.com/2p8zx4dp ) and published one research paper ( https://tinyurl.com/mtyzmh79 ). Couple of more patents are under review.
• 4. I currently own project to power our CRM and Marketing product with LLMs. Owning includes architecting, estimating product feasibility, roadmap and vision for next year and execution.
• 5. Worked closely with product team to iterate on mocks, user experience and defining success metric apart from ensuring engineering architecture and ML models were built in scalable and accurate way.
• 6. Projects completed include deduplication feature ( https://tinyurl.com/mrxd7hnj ), deal insight ( https://tinyurl.com/yru6yxmv ), send time optimisation ( https://tinyurl.com/bdh5xcux ), deal forecasting and product recommendation (WIP), RFM ( Recency, frequency, monetary).
• 7. Taken a broader role since July 2023 to manage product backend, frontend and QA engineering apart from ML team.

Apr 2020 – Mar 2022 · 1 yr 11 mos

Nov 2018 – Mar 2020 · 1 yr 4 mos

Jul 2018 – Oct 2018 · 3 mos

Dec 2016 – Jun 2018 · 1 yr 6 mos

• o Product
•  Lead team of data scientist in designing and conceptualising AI systems
•  Built various packages to ease data exploration, signal detection and modelling
•  Worked in area of deep learning to build time series forecasting and time series clustering model using feed forward neural nets and variational autoencoder models respectively.
•  I was instrumental in coming up with version 1.0 of product which helped raise series B round.
•  Different modules built were used in various engagements in area of demand forecasting, predictive maintenance and energy optimisation
• o Client engagement
•  Lead the team in India to build an AI engine that would cater to a large distributor of US to solve their demand forecasting problem.
•  Experimented with several forecasting techniques like XGBoost, Facebook prophet and ARIMA
•  Explored use of variables which influence sales in short and long-term

Apr 2016 – Nov 2016 · 7 mos

• Sherpa (AI Engine)
• We are working on products which would help marketers build smarter campaigns to reach to their customers using machine learning and customer behavioural data. We are trying to build a self-learning system which improves over time using reinforcement learning.
• Also, lead a team to build lambda architecture framework consisting of data lake and real time analytics system using logstash, kafka, apache samza and mongodb. Scale the system to process 200 GB of data in a day with stress on minimal devops, cost and high availability.

Nov 2014 – Mar 2016 · 1 yr 4 mos

• I was amongst the founding team at Moengage. I worked towards scaling the backend system here. Some of the highlights of my work are:
• 1. Scaled the tech stack to support 80+ clients tracking more than 12 billion events per month.
• 2. Re-architected initial monolithic architecture to a lambda architecture consisting of stream processing and building data lake.
• 3. From being an individual contributor to mentoring 20+ developers.
• 4. Architected backend system for various products at Moengage like campaign analytics, event tracking, smarttriggers etc.
• 5. Working at different engineering layers like infrastructure, database management, DevOps etc.
• 6. Hiring around 10+ backend engineers.
• Various technologies I worked on during this time:
• 1. MongoDB for storing unstructured and meta data of users, devices, campaigns etc.
• 2. We are deployed completely over Amazon Web Services. We rely heavily on various services like ELB, EC2, Auto Scaling, Elasticache, SQS, S3, Cloudwatch, Route53 for our products, monitoring and alerting systems
• 3. For various DevOps function we use NewRelic, Pager duty and Elastic Container Services
• 4. Backend application development is done on python (pyramid framework) and java
• 5. Some of our major products are powered by Elasticsearch.

Dec 2012 – Nov 2014 · 1 yr 11 mos · Bangalore

• o Predictive modelling (Data science team)
•  Architected and developed state-of-art machine learning system including training and real time prediction for click through rate prediction.
•  Training and data crunching extensively used Hadoop on Elastic MapReduce (Amazon service)
•  Explored various techniques including regression, decision trees, clustering for CTR prediction.
•  Deployed logistic regression model with online updates. Models learned and updated on hourly basis. Algorithms/techniques worked on stochastic gradient decent, BFGS, bootstrapping, boosting, logistic regression, explore-exploit ( epsilon greedy, softmax ), thompson sampling.Tools used Vowpal Wabbit, Scikit, H20.
•  Scaled system for a through-put of 30000 request/second with max latency of 5 millisecond on 8 core box.
• o Analytics platform
•  Built a real-time, fault tolerant analytics platform using Elasticsearch, logstash and lumberjack
•  100 mb of data generated each minute was made searchable with a lag of 15 seconds.
• o Demand side platform (DSP)
•  Architected and deployed cluster over AWS handling 40000 request/sec with max latency of 50 millisecond
•  Intensive use of caching and webserver tuning helped to achieve high throughput with low latencies per box
•  Deployed and tuned 8 node NoSQL database (Cassandra) to be used as a caching solution for user data
•  Proposed algorithms for real time bidding system based on various user and advertiser parameters
• o Complex event processing (CEP)
•  Mentoring an intern on CEP project which involves tracking and analyzing streams of real time data

Jul 2011 – Nov 2012 · 1 yr 4 mos · Bengaluru Area, India

• Core team member of Preflight Application Checker ( Migration tool for Solaris 10 to 11)
• Tool generates issues for migration of software from Solaris 10 to Solaris 11 using three scanner namely Binary, Source code and Runtime scanner.
• Design and implement new features from scratch for upcoming update of tool.
• Used several tools and third party APIs like XSLT, Apache FOP, Java, etc.
• Current release had 600+ download in a month.
• Design, development and support is managed by a small team of 5 people.
• White paper
• Co authored white paper "Red Hat Enterprise Linux to Oracle Solaris Porting Guide" which covers topic like system calls, commands, PAM, Device drivers, IPS, etc
• Oracle Database
• Working on performance related issues on the upcoming Oracle 12c. Using workload generation tools like swingbench, tpch, etc to study performance on virtual environment on unix systems specially Solaris 11.

Aug 2010 – Jul 2011 · 11 mos

• o Sentiment analysis and opinion mining using machine learning techniques.
• o Real time extraction of tweets from twitter and categorization of them into positive and negative sentiments. Model trained over Naïve Bayes classifier with an accuracy of 85%.

May 2010 – Jul 2010 · 2 mos

• Worked with the Development team of the advertising group. Studied various concepts useful in online advertising like finding communities of interest, recommendation systems, clustering, targeted advertising, use of social profile for advertising, etc.
• Generated a large set of related words given an input word.
• Eg. Coffee -> cappuccino, espresso, turkish coffee, beverage, etc
• This could be used to target users with related search keywords for online advertising.

May 2009 – Jul 2009 · 2 mos

• o Improved current version of Dialign program for sequence alignment by incorporating protein secondary structure information under guidance of Prof. Burkhard Morgenstern, Chair, Dept. of Bioinformatics.
• o Appreciated for quick learning abilities and for strong hold in C language.
• Publication:
• A.R. Subramanian, S. Hiran, R. Steinkamp, P. Meinicke, E. Corel, B.Morgenstern (2010)
• DIALIGN-TX and multiple protein alignment using secondary structure information at GOBICS
• Nucleic Acids Res.38, W19-W22

May 2008 – Jul 2008 · 2 mos

• Automation Opportunities in Credit Risk Process, Analytics team, Genpact (May 08 – Jun 08)
• o Automated credit risk calculation using VBA which saved 70% of total time required for the process.
• o Learned teamwork, professionalism and planning in India’s largest BPO organization.
• o Special appreciation by the Analytics team for my hard work and commitment.
• Knowledge Management, Genpact (May 08 – Jun 08)
• o Structured an online platform for Genpact where knowledge can be shared and distributed securely.
• o Interacted with people of various hierarchies to conceptualize the project.