Pratik Katte

[git repository]

This project aims to make tree-sequence analysis more accessible by integrating large language models (LLMs) with an interactive, natural language-based interface. Traditionally, analyzing tree-sequences—a powerful method in evolutionary biology and population genetics—requires programming expertise, limiting its use to computationally skilled researchers. By combining LLMs with Retrieval-Augmented Generation (RAG), the system can interpret user queries in plain language and generate accurate, executable code using the tskit library. Additionally, the project is developing a Taxonium-inspired visualization framework tailored for tree-sequences, enabling intuitive exploration of large-scale genomic data. Together, these tools simplify complex analyses and broaden access to cutting-edge evolutionary research.

This project is in collaboration with the Turakhia lab, UCSD. [git repository]

I’m developing an interactive, real-time visualization dashboard that integrates the output generated by the WEPP (Wastewater Epidemiology and Pathogen Prediction) pipeline, a powerful computational tool developed by the Turakhia Lab at UCSD. WEPP analyzes wastewater sequencing data to detect circulating viral haplotypes, quantify lineage abundances, and flag cryptic or unaccounted mutations—potential early indicators of emerging variants. In collaboration with the Turakhia Lab, my dashboard bridges the gap between complex genomic outputs and actionable public health insights by presenting this information through intuitive, interactive visual interfaces. It leverages tools like Taxonium for phylogenetic tree visualization and JBrowse for genome-level inspection of sequencing reads, allowing users to explore the relationship between detected haplotypes and observed mutations directly on the reference genome.

StructHunt won the first prize in the QBI Hackathon orgaizied at University of California San Fransisco.

We created a tool designed to track the publication of new research papers detailing integrative biomolecular structures in bioRxiv and medRxiv. This tool enables us to swiftly capture and incorporate this valuable new data into the RCSB Protein Data Bank. Within a remarkably short timeframe of 36 hours, our team swiftly gained insights into integrative biomolecular structures, various LLM techniques (special thanks to Lantern for their assistance in optimizing storage and retrieval of embeddings), and successfully launched a fully operational prototype. This prototype has been seamlessly integrated with GoogleDocs and email notifications, hosted in the AWS cloud environment.

XraySetu is a joint collaboration between Niramai Health Analytix, Indian Institute of Science and ARTPARK.

Covid-19 delta variant had a disastrous imapct on not only matropolitan cities but also even in rural parts of India. RT-PCR test which was heavily relied upon to diagnose covid-19 was resulting in misdiagnosis for the delta variant of Covid-19. Therfore, doctors started using chest x-ray scans to diagnose covid-19. In India, there is less than 1 radiologist for every million people. With the limited number of radiologists in the country, it was impossible for doctors in 2nd tier cities and rural areas to diagnose covid-19 using chest x-rays. Xraysetu played a huge role in allowing doctors to plan early intervention for their patients by simply taking a picture of their chest x-ray and sending it over via whatsapp.

The free Whatsapp based XraySetu service responds with a detailed report generated using our state of the art deep learning model.

The state of the art deep learning model is trained using multi-task learning on a combination of dataset from multiple sources which includes NIH and RSNA, etc. The model generates a report containing predictions for COVID-19 and 14 other lung abnormalities with interpretable semantic markings on chest x-ray. This helps doctors understand the severity of illness of their patients.

Xraysetu is widely covered in print and online media by NDTV, CNBC-TV18 News, Gadgets Now, Mint, Hindustan Times, Business Standard, Economic Times, ET Healthworld, KnockSense, Bangalore Mirror, Jagran, Zee News Hindi, and Prasar Bharti (video).

Niramai Fever Test project received research funding by CDC-Group.

The recent outbreak of covid-19 has brought a tremendous impact on the livelihood on the population. Community screening is the most important and primary aspect to reduce the spread of the virus in the cummunity. In order to detect people with covid-19 symptoms, we at Niramai, developed an AI based solution integrated with a thermal camera to measure temperature of a person and also detect if the person is wearing mask or not. We trained a light weight deep learning using thermal images for the task of face detection and mask detection.

Niramai Fever Test has been deployed in more than 100 locations which includes railyway stations, corporate tech parks, banks, etc. It has screened more than 10 lakhs of people in 2 years scince the deployment.

[More information][Video]

I lead the development of the Niramai Thermal Capture Tool, a desktop application used by hospital technicians for capturing thermal breast images and seamlessly uploading vital data to the Thermalytix website.

This tool was meticulously designed to empower technicians, allowing them to conduct essential preliminary checks before commencing image capture. Moreover, it plays a pivotal role in ensuring technicians adhere meticulously to Niramai's specialized screening protocol for diagnosing breast cancer.

By leveraging a sophisticated amalgamation of image processing algorithms and cutting-edge AI models, this tool reflects my vision to equip technicians with robust resources for conducting comprehensive initial assessments, revolutionizing the early detection and diagnosis of breast cancer.