My career has focused on the research, development, and application of AI/ML techniques, especially to robotics and biotechnology.

As a 20-year veteran in AI, I have witnessed the field’s evolution from early machine learning to statistical machine learning to deep learning, and now Generative and Agentic AI. As AI continues to evolve, my aim is not only to help engineer it but also guide its responsible application, helping safeguard against potential societal harms.

I have developed Halluciguard (https://github.com/prateekt/halluciguard/), a framework to QC AI agents. Halluciguard uses prompt-engineered checker agent ensembles to safeguard against LLM hallucinations, detect false claims made by any other agent, and to rigorously and automatically evaluate and grade work done by agents

In my recent role at Roche (2021-2025), I led efforts on Generative AI applied to Nanopore sequencing. I served as the primary technical subject matter expert on Diffusion Models (especially applied to video and image generation). Our team filed a patent application applying video generation techniques (like Sora) to Nanopore flow cell chemistry. I also spearheaded training data preparation for our Deep Learning genomic base caller models to call DNA bases (‘A’, ‘G’, ‘T’, ‘C’) from Nanopore sequencer signals.

My career spans AI innovations in robotics, computer vision, bioinformatics, natural language processing, and intelligent agents. Core application areas are biotech (including genomic sequencing, oncology, and psychiatric genetics) and robotics (machine learning for sensing and planning).

I hold a PhD in Robotics and Machine Learning from Carnegie Mellon University (2015), a postdoc in Bioinformatics / ML from UC San Diego (2019), and a B.S. with multi-agent systems and early neural networks research from the University of Southern California (2010).

I am excited by the future possibilities of upcoming technology. I led the Quantum Robotics Group (2015-2019) and coauthored a book on Quantum Robotics, setting a vision for how quantum computing advances will impact robotic intelligence. I have led Climate Tech Group discussions to combat global warming. In addition, I have over 10 years experience in leadership in the nonprofit space, building software for persons with disabilities and organizing hackathons globally.

Here is a summary of some of my previous research contributions:

  • Led the patent application for “Predictive Video Generation for Flow Cell Dynamics using Deep Learning Diffusion Models,” to simulate sequencer experiment videos from sequencer input parameters. Patent application filed internally at Roche (2024).
  • Led application of Deep Learning Diffusion Models to simulation of Nanopore sequencer signal traces (as images). Poster presented at Roche CSI Faire 2024.
  • Developed Fusion Selector, a method for combining multiple base caller predictions to generative higher-quality training labels for Roche’s on-GPU model. Consistently produced top-performing base caller models at Roche from 2021-2024. 
  • Worked on improving models of structural variation mutation rate in the human genome as a postdoc in the UCSD Sebat Lab. Poster published at American Society for Human Genetics (ASHG) 2018.
  • Worked on improving approaches to pathogenicity prediction of structural variation using machine learning. Poster published at Society for Neuroscience (SfN) 2018.
  • Organizer for NIPS Machine Learning for Health (ML4H) Workshop in December 2017. Workshop successfully aggregated clinicians and machine learning experts to discuss healthcare domain applications!
  • The Quantum Robotics Group I helped spearhead published a book in January 2017, Quantum Robotics: A Primer on Current Science and Future Perspectives. Book available on Amazon via Morgan Claypool Publishers!
  • Deep Learning consultant for UCSF Strigo lab exploring Deep Learning architectures for fMRI of pain representation task. Presented to Siemens Research Group, New Jersey in 2017!
  • Explored EMG and EEG signal processing as part of CyborgDistro plans to control multiple robot arms! CyborgDistro was featured on RaspPiPod, Adafruit, as a quarterfinalist in the Hackaday 2015 Prize Contest, and a winner of the Element 14 Halloween Build-A-Thon!
  • I defended my PhD Dissertation on Bayesian Aggregation at Robotics Institute, Carnegie Mellon University in July 2015. PhD Awarded!
  • Developed general machine learning algorithms for neural data and applications to experimental physics. Posters presented at Big Neuro Data and Aleph workshops at NIPS 2015.
  • Developed algorithms for Fault-Tolerant Radiological Threat Detection using Bayesian Reliability Models and Anomaly-Match Bayesian Aggregation (AM-BA) for Efficient Radiation Source Search. Posters presented at IEEE Nuclear Science Symposium 2015.
  • Lead developer of Lunar Tabs, an Intelligent, Accessible Guitar Tab Reader. Developed personalized difficulty classifiers for guitar learners using large amounts of guitar tab data online, enabling swarms of humans and robots to autonomously learn guitar. Project poster featured at AAAI 2015.
  • Building Biologically-Inspired Deep Learning algorithms that are more pragmatic, realistic, and believable by being more biologically plausible than backpropagation for tasks such as neural hammering. They can also have better optimization properties! Senior thesis on population-coded neural network controlled robots still pending publication after 6 years. When will the world be ready for Biologically-Plausible Deep Learning?