Jul 2021 – Present · 4 yrs 9 mos · Hyderabad, Telangana, India

• Helped scaled my client to spend 10 lakhs/month in ad spend, earning 5x ROI
• Built a Customer Data Platform using Google Tag Manager, Google Cloud, Neo4j
• Built an Identity resolution tech using Graph Data Science
• Built ML Attribution Models using PyTorch
• Built a Graph DB of first-party data using Neo4j
• Built data pipelines in Google Cloud to collect FB ads data into bigquery
• Built data pipelines in Google Cloud to collect CRM data into bigquery
• Built pipeline to collect user click event data in bigquery using Server Side Tracking

Sep 2019 – Apr 2021 · 1 yr 7 mos · Bengaluru, Karnataka, India

• Created a first-of-its-kind training module for building a complete no-code custom AI solution using Microsoft Azure platform using one's own first-party data
• Created a Google Spreadsheets plugin to fetch ML predictions from your Azure ML model directly into your sheets to make ML models accessible to non-technical professionals.
• Trained 140+ Digital Marketers to use their own first-party data and build custom AI models without coding
• Creating Applied AI Experts by making professionals solve real-world problems in their own line of work
• Built a holistic AI learning hub for non-coders to learn and apply AI through Courses, Bootcamps, 121 Coaching, Community
• Helped Karthik, a Digital Marketer increase his salary by 4x in 4 months
• Helped Aishwarya transition from Digital Marketing to Data Analyst and has seen a salary improvement of 200%
• Have been an AI Consultant for Wildcraft, SimpleCRM, SanjayShenoy.com, BIRetail in helping them build AI products
• Have been awarded Top 50 Chiefs 2019-20 by TapChief for completing a record number of gigs in 2019

Oct 2018 – Present · 7 yrs 6 mos

• Taught courses on AI, ML at BITS Pilani
• The course content includes
• Search Strategies e.g., A*
• Representation in Logic, Semantic Nets
• Monotonic Reasoning using Forward/Backward chaining
• Non monotonic Reasoning using Bayesian Nets and HMMs
• Learning Paradigms including Supervised, Unsupervised, Reinforcement
• Neural Networks and Deep Learning
• Optimization Models including Hill Climbing, Genetic Algorithms, Ant Colony Optimization, Particle Swarm Optimization
• Applications for Gaming including MinMax, Alpha Beta Pruning
• NLP Systems
• Understanding Fairness, Privacy and Interpretable ML

Mar 2016 – Apr 2019 · 3 yrs 1 mo · Bangalore

• Implemented Contextual Multi-armed Bandit algorithms to efficiently allocate doctors to user’s health queries. Observed a 60% reduction in doctor’s response time within 10 days and total answering doctors increased by 30%.
• Identified product’s limitation in fetching more doctor replies and developed an algorithm to dynamically match demand-supply. An impact of 50% more replies/day is observed.
• To ensure the quality of questions, I had built a spam filter which is an ensemble of five classifiers handling different subgroups of spam. It reduced the load on manual moderation from 40% to 5%.
• In order to boost SEO traffic, I had worked with Marketing team in creating 18k unique pages using data mining techniques.
• While posting a health query, user needs to select a speciality from 23 categories. This painful selection is reduced to 3 categories with 99% accuracy using NLP, ML.
• Collecting user event data is critical to understand users. As a data scientist, I took an initiative in defining consistent event language across the product flow and implemented the event definitions collaborating with different stakeholders.
• It needs a robust data pipeline to store/validate/model/use all user events across products. I had recommended efficient Data pipeline architecture, estimated costs, communicated design decisions to central engineering.
• Apart from my associated team, I had contributed to a different team’s client onboarding process by standardizing their client’s noisy text data to in-built medical ontology.
• Published a paper on “Understanding Medical free text: A Terminology driven approach” at COLING’s CompuTerm Workshop, 2016.
• Presented a talk titled “Adapting Bandit Algorithms to optimise user experience at Practo Consult” at Fifth Elephant 2017, India’s Leading Data Conference.
• An advisory to other Data Scientists within organization on model building/data science.

Feb 2015 – Feb 2016 · 1 yr · Mumbai Area, India

• Improved the existing property valuation tool using Random Forest with a bootstrapped interval prediction resulting in a 10% increment in hit ratio with a decrease of 14% in error interval.
• Predicting the lifetime leads for any live rental listing helps in optimizing ad campaigns. A gradient boosting predictive model is built to predict the lifetime leads using listing features with 86% accuracy.
• The data collectors fraudulent uploads are tough to identify because of high false positives. After few unsuccessful attempts, an image similarity approach is implemented using SIFT descriptors.
• Provided data-driven insights on rent and resale listings distribution across price segments. The analysis resulted in an actionable insight of collecting more listings in higher price segments.
• Provided real-estate investment insights by munging rent and resale data in exploring rental yields. This is communicated with a visual story and is published as a Housing blog post.
• A pilot task of professional market valuations required a representative sample of Mumbai resale properties. A stratified sampling along with CLT is used in estimating a sample size with 10% error.
• Mentored several DS/BI/Product Analysts with their NLP (sentiment analysis on User feedbacks, topic modeling on forums, matching building names), data munging and visualization tasks
• Using in-house data, insights are gathered on buying trends in Bangalore and is published as an article in Deccan Herald newspaper.
• Improving the user buy experience by providing Development status of localities using Machine Learning techniques.

Jun 2013 – Jan 2015 · 1 yr 7 mos · Pune

• Worked on building Machine Learning algorithms for addressing one-class text classification incorporating the NLP based structure information of the text as features.
• Alternatively, for the above task implemented a Point-wise Mutual Information (PMI) based approach to cluster related phrases, and these phrases are used in classifying the sentences into classes.
• Developed and maintained the categorisation of news articles into meaningful entities helping the clients to track topics of interest.

Sep 2012 – Jun 2013 · 9 mos

• Built statistical models for forecasting data from financial sector.
• Performed significance testing of the statistical models.
• Improved runtime 30 times by building the existing R statistical packages in c++ environment.

May 2010 – Aug 2012 · 2 yrs 3 mos · Hyderabad Area, India

• Represented IIIT-H at Text Analysis Conference 2009, in RTE track with a revamped system over our previous participation with an 18 percent improvement in accuracy.
• Pioneered a research project funded by Nokia in collaboration with IIT Bombay, the deliverables were reached well prior to the deadlines and influenced the IIT-B team to study the system developed.
• Developed the server-side of an Image search project and later took up the user-interface development when the front-end team had dropped out from the project.