Python · After surveying some existing open source NER solutions, we created a Named Entity Recognizer that : * would recognize entities which it… · More is trained upon * would recognize entities of n-gram. Enabling it recognize entities which had as much lengthy surface text as required
Its accuracy surpassed a few notable award winning open source and properitery alternatives that we had considered prior to coding our own
Natural Language Understading, Natural Language Processing, Graph Theory · Machines can understand keywords well. But they have hard time understanding… · More the important but suble difference in 'similar sentences'.
So the sentence : 2 large sub sandwich with extra mayo Is different than : two large submarines with no mayonnaise But yet similar in some features. Humans understand these differences and similarities of features innately.
The challenge was to make machine understand such differences and similarities of instruction. With such understanding a machine can perform user's commands better, return better search queries.
Here we made a trainable Sentence Vector generator which would take input of natural langauge text and output a sentence vector, which later can be used to help respond back to complicated search queries or commands requested by user.
While still in its infancy, this component enables Smarter.Codes to read and understand natural language, and its accuracy surpassed when tested independently against Wit.AI's trained models
GPU Computing, High-Performance Computing, Graph Databases · A common problem in Machine Learning algorithms is that they depend on Graph Databases and… · More require traversing in millions (if not billions) of graph nodes before giving any result. Such transversals often tend to slow down the ML algorithms and exhaust the physical resources of server machine pretty quickly
Our Solution: After benchmarking a dozen graph data Libraries and Database servers, we ended up creating a Graph Storage system from scratch in C++ which * supported write & read of billions of graph relations * utilized myriad hardware within machine including GPU, RAM, SSD and Hardisk to optimize on space and speed * operated on multiple nodes of interconnected machines
To optimize for memory space, some Graph Cut algorithms were implemented that would slice the graph in smaller portions while still keeping graph transversal fast on an array of GPUs, RAMs, SSDs and Physical Machines interconnected together.
40% more sales with AI for your Ecommerce store. Tag products automatically & create a powerful search that raises your conversions upto 1.8… · More times
Smarter.Codes achieves superior results by creating a hierarchical relationship of the catalog products and entities. Then, the database is trained to understand the associations one entity may have with the other. For e.g Apparels can enclose Shirts, T-Shirts, Jeans, Jackets etc.
1. 15+ years in Big data, Graph Theory, Metaphysics and Web crawlers .
2. 5th generation programming theories appreciated by the community.
3. Developed NLP based Market Analysis tool that gathered 36,000
4. Ran a profitable software company for 12+ years
5. Coded self driven toy cars & home automation as science projects
6. Honored with the Youngest Entrepreneur Award in 2008-2009.