Project Overview

  • This was the research component of OneNLP Stack library, which is IBM's in house solution for domain adaptive and language adaptive NLP. Work on this started during my internship and extended for 8 months during my full time role, before I moved on to Watson Health.
  • I worked specifically on the semantic components of this library, helping NLP systems differentiate the nuances in meaning of the same word, when used across domains and contexts
  • This work involved building on top of legacy code to enable an ML model (kNN) that assigns predicate sense and argument roles to tokens in a sentence
  • This required in depth understanding of language constructs and frameworks such as POS, semantic roles, framenets, verbnets and other computational linguistics topics
  • This was also my first time working on large scale enterprise systems and learning how to deliver innovation to existing systems
  • Work on this has provided me a US Patent Office granted patent
  • Additionally an API tool version of this project called SystemT was also presented in KDD 2019

Skills

Learning to build adaptive ML systems

Often times the off-the-shelf ML tools we use are garnered towards generic domains and don't exactly apply well to specialty domains or low-resource languages. Exposure to this project allowed me to develop a systematic approach to addressing this problem in the industry and coming up with solutions to it through research, domain understanding and a data-first thought process

Understanding and being able to control ML systems

Exposure to this large scale project taught me how to be responsible for the ML features delivered. Rather than saying it's a black box, how do we best explain it to enterprise clients who rely on our systems to make critical decisions. This starts with understanding the data, process and intentions of the system to the core. I was able to develop that level of control here, rather than going behind every new flashy technology.

Working cross departments and functionalities

Being part of a much bigger thing, I had the opportunity to learn to work with and across departments such as research, engineering, product evangelists and client engagement. This was also when I developed my networking skills reflecting symbiosis across the global research labs, because it truly does take synergy and co-operation to run a huge project smoothly.