Naveen Arunachalam AI Research Fellow, Long Term Future Fund Email: narunach AT alum DOT mit DOT edu GitHub: @naveenarun

I am currently an AI Research Fellow at UnSearch (Understanding Search in Transformers), a multi-university center advancing fundamental AI research, supported by the Long Term Future Fund. My work focuses on mechanistic interpretability, human guidance, and fine control of transformer AI model architectures, which have a wide range of applications, including LLM-based design of novel materials, proteins, and therapeutics. I completed my PhD at MIT in the Kulik Group, supported by the NSF, DARPA, and the Office of Naval Research. Personal Information Ph.D. (MIT) 2023; B.S. (Caltech) 2018 Teaching Fall, 2020, 10.585: Engineering Nanotechnology (TA) Spring, 2018, ChE 105: Dynamics and Control of Chemical Systems (TA) Winter, 2018, ChE 103b: Transport Phenomena (TA) Fall, 2017, ChE 15: Introduction to Chemical Engineering Computation (TA) Spring, 2017, ACM 95b/100b: Applied and Computational Mathematics (TA) Fall, 2016, ChE 62: Separation Processes (TA) Winter, 2016, Ch1b: General Chemistry (TA) Research My broad goal is to improve the design of novel matter. Within chemistry, I am specifically interested in advancing AI- and ML- accelerated virtual screening, whereby molecules of interest are rapidly discovered on demand based on current needs. I have worked on several areas in the design of novel matter, including: Mathematically calculating the properties of novel molecules using GPUs Database [web] of >100k molecules and their properties Inferring trends and design principles from chemical data [web] Accelerating design by using machine learning to calculate properties in milliseconds (rather than hours, in the case of GPU calculations) [pdf] Reducing the number of GPU calculations needed to train machine learning-based tools [web, web] Helping humans interactively co-create new molecules with machine learning-based tools MolSimplify Lite [web] - design tool for transition metal complexes MOFSimplify [web, pdf] - design tool for metal-organic frameworks Helping humans guide state-of-the-art AI model architectures (transformers), which are used in molecular design [web]