Our Mission: To Democratize Medical AI Research

LangChain is an orchestration framework that works in the backend to process user input, retrieve data from an external knowledge base, construct prompts and interact with the LLM.

Streamlit manages the frontend where it creates the user-facing web interface and controls the user interactions with the elements like text boxes, buttons.

Reference: Streamlit tutorial

AI for personalized medicine helps to achieve treatment response optimization and improve drug response prediction. It takes into consideration the unique characteristics of each patient at molecular, physiological, ecological and behavioral levels. It is quite the opposite of the one-size-fits-all approach, i.e., relying solely on clinical guidelines after determining the disease stage.

Reference: Journal paper

Structured data is organized in fixed fields like tables or spreadsheets, such as lab results; unstructured data includes free text like clinical notes, voice recordings or images.

The NITI Aayog (National Institution for Transforming India) is the premier policy think tank of the Government of India. In 2018, it released the "National Strategy for Artificial Intelligence," which outlined a vision for using AI to drive economic growth and social inclusion, with healthcare being one of the five designated sectors for application.

Why others are wrong:

NASSCOM is a trade association for the Indian IT industry.

C-DAC is an R&D organization focused on advanced computing.

ICMR is the apex body for biomedical research. While all are involved in technology and health, NITI Aayog is responsible for the national strategy paper.

Reference: NITI Aayog. (2018). "National Strategy for Artificial Intelligence."

Healthcare data comes from diverse sources and formats, making standardization critical but challenging.

Data abundance is an opportunity, not a challenge.

Missing data issue is a derivative of data standardization issue as different source collect different types of data leading to sparse datasets when aggregated together. For example, for LRRK2 Parkinson Disease cohort, data was collected from 10 hospitals or medical centers. Some centers did not collect genomic data, while others did not collect imaging data.

Data provenance captures the history and origin of a data point, tracking the changes made over course of time. In healthcare, changes made to patient data is heavily regulated and usually, comes from a single medical center without intermediaries. This is an important issue which is currently solved.

Ambient AI systems help clinicians by automatically capturing clinical conversations and documentation, reducing administrative burden.

For example, imagine a fully autonomous, ambient device that uses a host of physical sensors to monitor a patient in an ICU. One that might use thermal scanning to monitor patient temperature, or acoustic sensors that could detect cardiac arrhythmia, or radar sensors to monitor even the most minute respiratory changes. (Source: DeepScribe company blog)

DeepScribe is a company that is leveraging ambient AI to automate documentation – allowing clinicians to focus on providing care and increasing patient interactions, leading to better patient outcomes. Learn more at their website.

Ambience Healthcare is a company that provides patient-focused Ambient AI services such as enhancing specialty workflows, improving documentation accuracy, and optimizing billing, integrated with the EHR System. Learn more from their website.

Dimensionality reduction is a technique used in machine learning to transform a dataset with many features (dimensions) into a dataset with fewer features while preserving the essential properties of the original data. Learn more at this blog.

It is typically used for data visualization. For example, you will find tSNE plots in representation learning papers where they show meaning clustering of learnt embeddings in terms of one or two distinctive features like gender, article subjects/topics.

Clustering groups data points based on similarity without labels, while classification and regression require labeled data.

Reinforcement learning differs from supervised and unsupervised learning where it involves an agent interacting with an environment via a set of actions and learning based on feedback or reward.  Unsupervised learning solely relies on understanding the data distribution and does not depend on any reward or feedback.

The Ayushman Bharat Digital Mission (ABDM) aims to create a national digital health ecosystem. A core component is the creation of Ayushman Bharat Health Accounts (ABHA) and the standardization of electronic health records (EHRs). This provides a foundational layer of structured, interoperable health data, which is essential for training and deploying effective and large-scale AI models in the Indian context.

Reference: Official website of ABDM

The proposed list of AI Governance principles as outlined in Page 4 of the report, which is currently (June 2025) at the public consultation stage are:

1. Transparency
2. Accountability
3. Safety, reliability & robustness
4. Privacy & security
5. Fairness & non-discrimination
6. Human-centered values & 'do not harm'
7. Inclusive & sustainable innovation
8. Digital by design governance

Reference: Sub-committee report PDF