Can AI Think Like an Engineer?

Why It Matters

Improving interpretability and deploying multimodal AI in autonomous labs could reduce risk and speed up scientific discovery across medicine, materials and security-sensitive research, making AI both safer and more practically useful in high-stakes environments.

Summary

Akshita Kamsali, a fifth-year PhD candidate in Purdue’s Elmore Family School of ECE, studies deep learning with a focus on computer vision and natural language processing, emphasizing interpretability to prevent spurious or unsafe model decisions in sensitive domains. Her work evolved from computational and experimental optics into AI around 2021, and last summer she applied large language and multimodal models at Lawrence Livermore National Laboratory to help power “self-driving” laboratories that autonomously run and adapt experiments. She explains computer vision as machines interpreting images and NLP as token prediction or ‘finishing sentences,’ and describes multimodal systems that fuse vision, text, audio, and sensor data to make integrated decisions. Kamsali frames these systems as tools to accelerate discovery and assist scientists rather than replace them.

Original Description

Explaining Trustworthy AI & Self-Driving Labs with Purdue PhD Candidate Akshita Kamsali

In this episode of Engineering Innovations from Purdue’s Elmore Family School of Electrical and Computer Engineering, host Kristin Malavenda interviews fifth-year PhD candidate Akshita Kamsali, who works with Professor Avi Kok on deep learning with a focus on computer vision and natural language processing. Akshita shares how her engineering background and Purdue exchange experience led her to pursue a PhD at Purdue, and how her research evolved from optics to AI. She explains her work on understanding and improving the reliability of deep learning models used in sensitive areas such as medicine and security, and discusses her Lawrence Livermore National Laboratory internship exploring large language models for multimodal “self-driving labs,” emphasizing challenges such as safety, security, transparent object sensing, accuracy, and efficiency. Akshita also reflects on the realities of the PhD journey, advising, and career plans.

00:00 Welcome and Guest Intro

01:00 Early Engineering Influences

02:07 Choosing Purdue for PhD

02:43 From Optics to AI Research

03:33 Deep Learning Explainability

04:24 Computer Vision and NLP Basics

05:46 Self Driving Labs Explained

06:36 Multimodal Models in the Lab

10:21 Safety Security and Reliability

12:20 Why Multimodal Data Matters

14:10 Life in a Fifth Year PhD

16:52 Advisor Style and Lab Culture

17:57 Who Should Do a PhD

19:42 Career Plans After Graduation

20:27 Free Time Walks and Cooking

22:11 Closing Thanks and Subscribe

Purdue ECE: http://engineering.purdue.edu/ECE

Facebook: http://facebook.com/purdueece

X: https://x.com/PurdueECE

Instagram: https://instagram.com/purdue.ece

LinkedIn: https://www.linkedin.com/company/purdue-ece

Purdue University's Elmore Family School of Electrical and Computer Engineering, founded in 1888, is one of the largest ECE departments in the nation and is consistently ranked among the best in the country.