Srinivasan Rangarajan

VueMail: Realtime Dynamic Image Compositing Service

by

VueMail is a image service built on top of Vue.ai Platform. This service can dynamically composite images on the fly when an email newsletter is opened. This is used by Ecommerce stores when sending large email newsletter campaigns.

One of the client had to send email newsletters to 8+ million subscribers with recommendations based on their browsing history and personal preferences. In the traditional approach, they would have to make 8 Million API calls to our recommendation API and embed the images for those products and send them. The problems with that is:

  • Expensive to do 8 Million APIs everyday, when the number of subscribers who will open it will be less than a Million.
  • Stock information will become stale for subscribers who open later
  • Email is not personalized based on the subscriber’s recent activity on the site.

The solution is to build a service which can generate the images on the fly whenever an email is opened. This VueMail service will be hit by millions of requests for images and it has to make sure to synchronize the calls to the backend services and return back the images of the product, dynamically composited with the latest price and stock information.

This service is now running in production with almost no maintenance and is capable of handling tens of thousands of image requests per second.

Tech: Python, Django, Pillow, Redis, Docker, AWS ECS.

Vue: AI Platform for Automating the Retail domain

by

Vue.ai is an Artificial Intelligence Suite of products and tools which helps automating the Retail domain. The product which started out as a simple recommendation engine widget embedded on ecommerce websites, is now an extensive AI based Software Platform which can automate various parts of a retail – right from the website and marketing channels to the backend operations.

There are different stages in the pipeline – Ingestion of the catalog, routing the catalog information to the next stages, Processing of the images and converting them into high dimensional vectors, do a very fast search on all the processed images, and serve the results over a REST API.

Tech: Python, multiprocessing, Postgresql, Redis, Docker, Go, AWS.

SocialGist: Track & Analyze Social Media Conversations

by

Designed and Developed the architecture for the SocialGist.com product, capable of handling millions of streaming data from various social media websites, including Chinese Social Networks like Sina Weibo, Tencent, RenRen, etc. Once the data is fetched, it is passed through various enrichment processes like Named Entity Recognition and Sentiment Analysis after which it is passed to an ElasticSearch Cluster.

The customers of SocialGist has the option of able to download all the data in bulk or stream them in near real time. They can also download the raw unprocessed feed vs the enriched feed.

Tech: Python, Kafka, Cassandra, ElasticSearch, Solr.

Daily Deals and eCommerce Analysis

by

In Bonobo Analytics, I built this entire product from the ground up. This product fetched daily updates of products sold on Daily Deals sites like Groupon, LivingSocial, etc., and source products from eCommerce websites. Using all this data, it analyzes and recommends best-selling products and deals best suited for that particular region/season.

This was a custom developed Django based product, which leveraged image recognition, natural language based processing and other data analytics to give insights into the performance and profitability of different group-buying sites. This offered valuable information to investors and shareholders in the eCommerce companies.

Tech: Python, Django, MongoDB, Postgresql, ElasticSearch, AngularJS, AWS, Vagrant.

SpotMaze: Private Search Engine for all your Personal Digital Data

by

SpotMaze v0.1 was built in about 48 hours in the Yahoo Open Hack Day and eventually became the product for the company (Cruns) I founded. The product was a secure and private search engine for all of your Personal Digital Data.

Everything from documents on your disk to cloud, photos, videos, social network updates, browsing history, etc., were indexed on a separate search engine for each user. The user had complete control over what data to be indexed and was able to delete them immediately.

Tech: Python, Flask, ElasticSearch, Postgresql, Apache Tika.

Voice of the Market: Sentiment Analysis Engine

by

Voice of the Market (VOOM) is a sentiment/opinion analysis product which can analyse how positive/neutral/negative a consumer product. It uses the scraped data fetched from numerous websites like review sites, discussion forums, social media, etc.

I built the first version which used a customisable heuristics engine along with a ML classifier. The analysis was done for each sentence which allowed adding numerous rules to use proximity of words to identify whether the user is speaking about two different sentiments about two features in the same sentence.

This product has support for multiple languages by having a pluggable architecture for the POS Tagger, rules engine and the models. It was tested against English, Polish and Korean.

Tech: Python 2.6, Flask, Postgresql.

KnowledgeMate: Automated Newsletter Creation product

by

KnowledgeMate was a newsletter creation tool, which used the data fetched from different review sites, forums, social networks, etc., and uses the entities extracted from them and presents in an easy to filter tool. Once the user has created the filters using plain text search queries, the different posts will be directed into different folders/tags based on the rules.

Based on the frequency, the user will receive daily or weekly emails based on his/her selection of tags. This tool was created at a time when ElasticSearch or the Percolator Query wasn’t built yet.

Tech: Python 2.6, Postgresql, Pyparsing, Lucene.