Data Software Developer

Ravi K. Jagannathan

ac23t4@r.postjobfree.com 408-***-****

Areas of expertise

Building machine learning products with emphasis on yield management, user acquisition, pricing and modeling. Industry experience in display, mobile advertising and apps.

Education

2000 MBA University of California, Berkeley - Haas School of Business

1992 ME Computer Science - Indian Institute of Science

1988 BSc, Mathematics - University of Madras

Vungle, Architect, Data science 04/2016 to Now

Vungle is a $300 MM/year mobile video ad network. My team designs the data science strategy including project roadmap, team hiring and business plan projections.

Currently, I am a fulltime architect in the product management group.

My work is designing on two industry-first projects involving 15+ engineers. The first is related to automatically optimizing creatives and was profiled in the prestigious Mary Meeker’s 2017 state of internet report. The second is related to monetizing full life-cycle of user, is not yet announced.

Applovin, Director, Analytics and Data science 05/2014 to 3/2016

Applovin is a mobile ad network with more than $ 300 M / year revenue. I managed the yield on the network with a team of 5 data scientists and 3 engineers, and reported to the CTO & CEO.

My individual work was in feature extraction, modeling result evaluations, and reporting on revenue gains gains. The platform we used was Vertica, Tableau and custom code in python.

•Targeting ads to 2B users is not manageable . Created an algorithm to segment them into “persona verticals” based on the mobile apps they use. Examples of personas are traveller, shopper, parent etc. This gave 3-5% lift, on the $100M revenue.

•It costs the advertiser significant amount to experimentally spend to learn on 800K publishers individually. We developed two machine-learning models for fine segmentation of apps space based on their descriptions and predicted gender of the user.Instead of using budget to test the expected returns of each ad against each app individually, it was enough to learn across a segment of apps . The methods used in the segmentation were Stanford NLP package, concept modeling, nearest neighbor clustering for segmentations, and logistic regression for classification of new apps.

•Publisher fraud using bots prevented the business from scaling through a self serve platform. Developed bot detection framework, that used a form of logistic regression to flag pubs with unusual distributions of IP addresses, locations, devices and other factors. Eliminating manual review increased the number of publishers signing up into the system by over 200%.

Applovin was acquired for $1.5B in 2017.

Million Levers & Apple, Architect user models 05/2013 to 05/2014

Apple’s had rich data about users, but it was siloed, and not usable for ads. Our project combined data sources from iAd and iTunes Apple business units to construct a customer models for targeting of ads and product recommendations. Led the outsourced team of engineers and data scientists across multiple technologies data science, Hadoop and Vertica. We used k-means clustering to segment users based on the apps they purchased.

Million Levers & GSN, Bash Gaming studio 11/2012 to 5/2014

In freemium games, it is very important to identify the 1% of paying users early, before they pay, so that they can be retained and incentivized to stay longer. To solve this problem, I co-founded Million Levers, a prescriptive analytics service that uses machine learning. Bash Gaming as the first beta customer. The model learns playing patterns of successful users and applies them to incoming users to pick the best players. We improved revenue modeling the probability of a user returning, and prescribing offers to retain the user longer, and to acquire better quality of users. The retention improved by over 5%

•Took an undefined fuzzy paper concept, and incrementally improved it until we acquired 3 paying customers.

•In limited funding, I hired a world class team of 2 PhDs, and 2 engineers by providing challenging problems as the motivation.

•Actively managed the business from raising angel funding to gracefully selling the technology.

The technology used a suite of classifications methods such as SVM, logistic regression and gradient boosted decision trees to select features, and then score a user as a payer/non-payer. The platform used a web server with a Vertica backend, for real time responses at scale.

Bash Gaming was acquired by GSN, a large media company, for over $150M

GSN, Idle Games Studio, Head of BI 9/2011 to 11/2012

Idle Games is publisher of Fresh Deck Poker, Idle worship and other creative games on Facebook. GSN is a large media company that owns brands such as wheel of fortune.

•Third party analytics solutions do not provide detailed insights into the product, as their dashboards are only the common to everyone in the industry. I replaced third party solutions with a more powerful in house solution made of Python, Tableau + Vertica + Hadoop stack. This solution empowered analysts to do ad hoc data analysis.

•As there was no team, and little time in a startup environment, I built the first version myself and then hired a team of a 4, DBA, two data analysts and one data engineer to scale the solution to hundreds of charts and trained biz users to create their own reports via self serve.

•Improved pricing and monetization by differential pricing by modeling users, and game play behavior. The python based scikit-learn was used for classifying users.

Idle Games was acquired by GSN for unspecified sum.

RockYou, Sr. Director, Yield Management and Analytics 2/2010 to 9/2011

RockYou has a unique business model of publishing social games with over a100M users, as well as running a large ad network of 1B impressions/day that displays ads inside Facebook applications. This allows RockYou to get a large amount of users for their games business from ad network business.

•Brought positive returns to user acquisition, quality and profit margins on the network through complex analytics managing a $2M/month budget. I was given additional responsibility of maintaining analytics for entire company including the games business.

•The two businesses RockYou ad network and RockYou games were cannibalizing each other.Too many in house ads would lower the ad revenue, too less would lower digital goods revenue. Developed a sophisticated analytics platform to consume right balance of ad impressions by accurately tying the purchase of clicks to dollar revenue. This reduced the advertising costs by over 1 million dollars a month, and achieved positive ROI on advertising costs,

•Every hour or day serving a suboptimal feature causes drop off in the game, and loss of revenue. Created a set of user engagement metrics and real time dashboards that allowed product managers to change content within 15 minutes of release if the user experience turned out to be unexpected.

•Rock you acquired several private companies, but the challenge was to verify the revenue stated by the acquiree. Evaluated opportunities, for M&A team, by analyzing the potential traffic for quality, monetization potential. Stated revenue was adjusted for fraud, bots, and revenue from internal or unattributed sources.

•Reported to the CRO with dotted line to CEO. Managed 8 direct and 10 indirect

The platform used was MySql, Hadoop, and Scipy for statistical packages

Nominum, Principal Architect, Search 1/2009 to 2/2010

Nominum provides DNS services to over 100 million users. Skye Search monetizes mistyped domain names by showing a search page with suggestions for the typo, as well as the advertisements.

•Increased revenue by over 10% by developing a data mining system to study traffic patterns and optimizing page design and search results based on the data.

•Page not found Error pages are uninteresting. Developed a concept of social browsing experience, using popularity and similarity of DNS requests. This was developed using Hive and Hadoop.

•Promoted to Director of product management to find new revenue streams. Identified DNS based CDN, and data services as new opportunities.

Nominum subsequently exited the data service business, due to privacy concerns from ISP partners

Yahoo, Architect, Search Marketplace Design 11/2006 to 5/2008

The market place team designs the complex rules of the search ad auctions, to maximize revenue to Yahoo.

•Optimizations of revenue were based on metrics averaging across the 1 million advertiser and 5 million keywords. Drastically reduced the complexity by developing a graph based clustering algorithm to create 8K verticals that could be modeled and optimized separately.

•As Yahoo resells clicks from various sources, the advertiser ROI is not consistent. I mitigated this effect by pricing the partner traffic using statistical models, and developed measurement mechanisms to determine the ensuing marketplace results.

•Increased Marketplace revenue by using a Linear Program to optimize for budgeted advertisers.

All development was on Hadoop, which was later open sourced.

Adobe, (Efficient Frontier), Senior Software Manager 1/2003 – 11/2006

I joined as the first hire in Engineering, and built a software platform, that applies Wall Street portfolio trading algorithms to efficiently buy $400 million of advertising per year from Search Engines. I had engineering responsibilities for two areas, the optimizer for advertising ROI and software scalability.

•Took the product into several market verticals such as loans and auto by creating specialized statistical learning algorithms for promotions, seasonality, volatile search markets, very sparse data, and advertiser selection of budget risk tolerance.

•The statistical algorithms were controlled by several parameters which required special knowledge to change. This prevented the growth of the operations team. I created a natural language interface to control statistical algorithms. Currently in process of Filing two provisional patents for the same.

•The advertisers had to spend substantial money to learn on performance of keywords. I introduced a concept of co-operative where they could exchange data in a co-operate to compete manner.

•Enabled the company to enter the ad agency market by designing a third party controls for the platform. The agency product got about $40 K in revenue per month.

•Reduced the operational costs of the company by 50% by robust automation and redesigning the data model.

•Enabled the company to scale to hundreds of customers by reducing the run time of the algorithms from 3 days to 2 hours by using in-memory databases, caching techniques and architecting the product as layers of self-contained processes.

All algorithms were coded using python on a Postgres backend.

Prior 10 years Experience as a UNIX /Linux kernel software developer for Storage and File Systems at IBM, Stratus computer and Cisco Systems. This was developing low level software such as device drivers, memory managers and firmware interfaces

ac23t4@r.postjobfree.com 408-***-****

Contact this candidate