Hi, I'm Tara K. Jain.

I am an Engineer specialized in AI.

Previously, I was invited to attend a special in-person Educational Program for the “Gifted Youth” at Stanford University, California in Cosmology.

I was awarded two bachelor’s degrees simultaneously - B.S. in Electrical and Computer Engineering and B.S. in Atmospheric Sciences both at the University of Illinois at Urbana-Champaign where I was advised by Erhan Kudeki and Eric Snodgrass.

Subsequently I worked as an engineer at Google in Mountain View where I built models and systems to solve problems at the intersection of Science, technology and data (GoogleX).

I am currently pursuing a master’s in AI with a focus on NLP + RLHF in the UK.


Recent / Highlights

“Deep Awareness / Deep empathy”: A system design for AI systems to respond to humans’ personal questions [Active]
This study analyzes conversation patterns to identify when users of AI assistants may be experiencing emotional distress and proposes a system to handle such cases. The approach prioritizes privacy and protection while facilitating human connection.
I led the facilitation of experiments at a one-of-a-kind radar field in Peru to improve ECMWF forecasts for stratospheric winds - modeling patterns in the data for Project Loon to enable high-altitude balloon navigation and recovery.

Prior

Linguistic and Cultural Bias Analysis in Large Language Models

Linguistic and Cultural Bias Analysis in Large Language Models

Tara K. Jain

This project explores how linguistic and cultural biases present in training corpora shape transformer-based LLMs default "values" during inference. By examining the influence of pre-training biases, we investigate how models might reflect or reinforce cultural norms without deliberate intent. The study illuminates hidden priors embedded in language models and their implications for fair representation and multicultural inclusivity in AI systems.

Automating defect detection for cable manufacturing with Deep Learning

Automating defect detection for cable manufacturing with Deep Learning

Designed and trained novel convolutional neural networks (CNNs) using PyTorch for object detection and defect classification in cables. Focused on model accuracy, real-time inference, and error detection. Applied deep learning to automate visual inspection and improve quality control.

Technical infrastructure for billing enterprise Cloud customers at scale as part of GCP

Technical infrastructure for billing enterprise Cloud customers at scale as part of GCP

Google

A robust subscription management system for GCP services built with a gRPC/Protobuf based architecture. This project included integrations with external vendors to fulfill custom cloud infrastructure requirements for enterprise clients, providing scalable billing solutions that significantly enhanced internal service reliability improving the customer experience.

eVNA API Development & Data-Driven Optimizations as part of the Pixel Phones Team

eVNA API Development & Data-Driven Optimizations as part of the Pixel Phones Team

Google

The work on the Pixel Phones team involved design and implementation of a novel 'eVNA' API in collaboration with industry partners such as STMicroelectronics and National Instruments. Analysis of data informed design optimizations and product development, ultimately enhancing connectivity and user experience for millions of Pixel phone users worldwide. The API has been included in all Pixel phones produced since its inception.

Calibration Algorithms & Manufacturing Data Analytics as part of Pixel buds

Calibration Algorithms & Manufacturing Data Analytics as part of Pixel buds

Google

The Pixel Buds project required implementation of complex calibration algorithms for critical technologies such as Bluetooth, where code must ensure performance and compliance with requirements that vary by country and region. Analyzing large-scale datasets proved crucial for improving product quality throughout its development.

Deep atmospheric analysis of Mesospheric Na Variability and its dependence on Geomagnetic and Solar Activity

Deep atmospheric analysis of Mesospheric Na Variability and its dependence on Geomagnetic and Solar Activity

Presented at American Geophysical Union, USA Indira Saladi Award for Outstanding Achievement. Youngest Principal Investigator at the Arecibo Observatory

This research presented at the American Geophysical Union utilized advanced resonance lidar systems at the Arecibo Observatory to analyze mesospheric sodium (Na) variability and its correlation with geomagnetic and solar activity. The data and models provided significant insights into seasonal and aperiodic variations in atmospheric composition.

High-Frequency data collection and analysis of the E region of the Ionosphere with the world's most powerful radar at Arecibo Observatory

High-Frequency data collection and analysis of the E region of the Ionosphere with the world's most powerful radar at Arecibo Observatory

Presented at American Geophysical Union, USA Max Hammond Grant for Outstanding Research in the Space Sciences

This project used radar data observed at the Arecibo Observatory to model how powerful radio signals can change the ionosphere, the research revealed new insights into the ionosphere and its behavior.


Posts / Tutorials

GPT models (such as GPT 3.5 onwards) don't always permit fine-tuning or adaptations with LoRA. However, system customization instructions are supported and can be used to make powerful prototypes. A demonstration is linked.
A comprehensive tutorial on developing topic classification models for NLP, featuring neural network implementations from scratch in NumPy to illustrate feedforward and backpropagation algorithms, as well as demonstrations using industry-standard packages such as PyTorch.