This was a highly scalable architecture that we built on a HF trading system where a news monitoring system (including capture, importance analysis and signal extraction using NLP) was deployed over an existing trading system which markedly improved stock selection and timing and helped get better returns.
Score news on economy based on relevance, extract features and correlate with other macro economic indicators. This aimed at explaining the economy and stress testing using the Smets - Wouters model.
This was aimed at augmenting video platform recommender by monitoring the activity on social media. For the purpose mainly Youtube , Twitch were used along with FB and Twitter.
Analysis into product reviews. The project aimed at identifying the key pain and gain points of a product for a customer based on reviews and also rank the reviews based on bias e.g. fake reviews, competitor reviews etc. Helps OEM to act on customer complaints.
Monitor google search, blogs, social media and identify the key topics under a theme, trending topics, emerging topics to be used in automated text generation based on the selected topics thus aiding SEO.
This was a project that mapped eCommerce products to social influencers and vice-versa. Social media (FB, Youtube, Twitch, Twitter, Path, Reddit, Instagram) was monitored for influencers and scored post accounting for engagement, following, fake or bot activity, etc.
An in house product aimed at diagnosis of automated diagnosis of skin conditions using millions of images trained on a Pytorch/tensor-flow/Keras framework using deep learning with CuDNN 7.1
Using CCTV feed detect diseases in a cannabis planataion at the onset and inform relevant personnel to act upon immediately so as to prevent outbreak of disease.
Sometime news appears in print media later than video news items. The project aimed at two aspects i.e 1. converting text ribbons in a news video to usable text to be used in NLP and 2. To convert the speech to text using a model trained on a 100K acoustic samples treated for accent.
Using image match a deep learning model was created to automatically suggest complementary pieces of attire and accessories. The model also accounted for local tastes and availability apart from obvious items like body shape and size, complexion, etc.
An extension of the above mentioned SEO project this was aimed to suggest images to be used based on the text descreiption or Alt-text on a website.
This was a smaller part of a larger project which aimed at finding the similar product across eCommerce sites and providing the best price possible. This part was used to find similar products where text match and pothers tests failed.