Tech Lead-LLMs, Senior Applied Research Scientist

• Jul 2023 - Present

Training backbone LLMs on Cerebras Wafer scale Engine.

Senior Applied ML Scientist

• Jun 2022 - Present

Enabling training of large language models Cerebras Wafer Scale Engine. Responsible for numerical analysis of training dynamics for long convergence training of LLMs.Developing stateful dataloaders to allow determinism between multiple training runs. Implementing SOTA AI algorithms and training techniques in Cerebras Stack.

Senior Member of Technical Staff

• Jun 2022 - Present

Working on ML applications for Wafer Scale Engine

Founder and CEO

• Dec 2018 - Present

Senior AI Research Engineer & Tech Lead

• Jan 2019 - May 2022

Deep learning research and development for Huawei Ascend chips. Built end to end conversational models for Ascend 310 inference accelerator. Built compression algorithms to improve memory utilization and speed of BERT, GPT. Managed the NLP team in terms of setting the technical goals, provide direction and provide guidance to the team. Currently focussing on improving the second order optimizers for deep learning models. Goal is to achieve acceleration on Ascend chips. Deep learning research and development for Huawei Ascend chips. Built end to end conversational models for Ascend 310 inference accelerator. Built compression algorithms to improve memory utilization and speed of BERT, GPT. Managed the NLP team in terms of setting the technical goals, provide direction and provide guidance to the team. Currently focussing on improving the second order optimizers for deep learning models. Goal is to achieve acceleration on Ascend chips.

Founder

• May 2017 - Mar 2020

In this meetup group we discuss the ideas behind AGI (Artificial General Intelligence) and emergence of potential Superintelligence both from philosophical and technical standpoints. We aim to play our part in the vital conversations taking place in tech industry regarding morality of artificially intelligent beings, and how can biological and machine intelligence can prosper together. https://www.meetup.com/Concerning-Machine-Intelligence-AI/ In this meetup group we discuss the ideas behind AGI (Artificial General Intelligence) and emergence of potential Superintelligence both from philosophical and technical standpoints. We aim to play our part in the vital conversations taking place in tech industry regarding morality of artificially intelligent beings, and how can biological and machine intelligence can prosper together. https://www.meetup.com/Concerning-Machine-Intelligence-AI/

Machine Learning Engineer

• Apr 2017 - Dec 2018

Working with Addictive’s Data Team to accelerate and improve the RTB (real time bidding) performance of in-app advertisements. Primary objective of my role is to parallelize design and migrate machine learning systems to big data ecosystem using Apache Spark and Hadoop. Designed, deployed and monitor gender, age prediction model that leverages user’s app-usage history to make relevant bids for mobile ad space. Developed geo-targeting system to target users in the proximity of a specific location using geo-spatial analysis. Piloted the task of laying down best practices to develop production ready code for big data systems. Currently working on mobile web analysis system to generate user interest profiles by analysing content of webpages visited by user using techniques like Named Entity Recognition, Word Vector Embeddings and Deep RNNs. Tools Used: Python, Scala, Aerospike, Redis, Git, MongoDB, Spark, Hadoop, Tensorflow. Show less

Data Scientist

• Nov 2016 - Apr 2017

Building smart e-commerce shopping platform using machine learning techniques. Designed scalable Product Categorization System to automatically place products uploaded by vendors in their designated category. Performed product classification based on name and description using NLP (natural language processing) techniques such as Topic Modelling and Information Retrieval. Working on building recommendation engine for Daily Grabs' e-commerce platform using data points from customers’ online behavior. Tools Used: Python, Scala, Spark, MongoDB, AWS EC2, AWS EMR, Git, nltk, SpaCy, scikit-learn Show less

Research Assistant

• Sep 2015 - Nov 2016

Worked on development of numerical algorithms for computational analysis of fluid flow. Implemented modified conjugate gradient based solver for partial differential equations arising in fluid dynamics. Tools used; Unix, C/C++, Python, Matlab Worked on development of numerical algorithms for computational analysis of fluid flow. Implemented modified conjugate gradient based solver for partial differential equations arising in fluid dynamics. Tools used; Unix, C/C++, Python, Matlab

Graduate Research Assistant

• Sep 2012 - Apr 2015

Proposed an inverse neural network based algorithm for solving inverse mapping problems arising in electrical engineering chip design tasks. Applied unsupervised learning (clustering algorithm) to partition unlabelled input data. Developed feedforward inverse neural network model to each cluster separately and combined each of the sub-models to generate a composite model used for final design. Used Quasi-Newton optimization for feedforward neural network weight optimization. Thesis can be found at: http://www.ruor.uottawa.ca/handle/10393/32249 Publication: Gosal, G. et. al, "Transmitarray Antenna Design Using Forward and Inverse Neural Network Modeling", IEEE Antenna and Wireless Propogation Letters, 2016 http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=7369941 Show less