Centre for Responsible Autonomous Systems in Healthcare

An interdisciplinary research studio where clinical expertise drives AI development.

Our Vision

A future where AI advances clinical care safely as AI systems become more capable and autonomous while keeping clinicians at the centre of every management decision for our patients.

Our Mission

To build the datasets, evaluation frameworks, and foundations that healthcare AI needs to be developed and deployed safely across the patients, workflows, and contexts where it will actually be used.

What the lab has built since founding.

10+

Most RSNA 2025 abstracts by any Indian healthcare AI lab

15+

Papers & abstracts at RSNA, MICCAI, NEJM AI, IJRI, KCR, AOCR since 2025

10+

Collaborating institutions across India, US, Europe, and Asia

90+

Clinicians, researchers, and engineers across CRASH Lab projects

How we started

Since starting the research group in April 2025, CRASH Lab has grown from a single research cohort into a nationwide effort spanning AI evaluation, data infrastructure, and clinician-AI collaboration. Our work is extremely fast-paced and is built around the thesis that healthcare AI should be evaluated by experts under real world scenarios. The timeline below traces what we've shipped since.

CRASH Lab Journey: The Milestone Timeline

April 2025

CRASH Lab Founded

CRASH Lab began as a small cohort of physicians, primarily radiologists, and engineers working together on radiology AI research projects for RSNA.

May-June 2025

Scaling the Research Studio

As our initial projects expanded, more physicians across specialties, medical students, researchers, psychologists, and engineers joined us. This helped CRASH Lab evolve from a radiology AI initiative into a broader healthcare AI research lab.

July 2025

Academic Validation

CRASH Lab's early work received academic validation through accepted research across healthcare AI and medical imaging venues.

Sept 2025

RadLE Benchmark Unveiled

We launched Radiology's Last Exam, a benchmark designed to test how frontier AI models perform against expert radiology reasoning.

Nov 2025

RSNA Research Milestone

We reached a major RSNA milestone with 11 accepted abstracts, reflecting our growing research output in radiology AI.

2026 and Beyond

National Research Collaborations

We began securing support from leading national bodies, including ANRF, DST, and the IndiaAI Mission, to advance responsible healthcare AI research.

Scaling the Research Studio

May-June 2025

RadLE Benchmark Unveiled

We launched Radiology's Last Exam, a benchmark designed to test how frontier AI models perform against expert radiology reasoning.

Sept 2025

National Research Collaborations

We began securing support from leading national bodies, including ANRF, DST, and the IndiaAI Mission, to advance responsible healthcare AI research.

2026 and Beyond

April 2025

CRASH Lab Founded

CRASH Lab began as a small cohort of physicians, primarily radiologists, and engineers working together on radiology AI research projects for RSNA.

July 2025

Academic Validation

CRASH Lab's early work received academic validation through accepted research across healthcare AI and medical imaging venues.

Nov 2025

RSNA Research Milestone

We reached a major RSNA milestone with 11 accepted abstracts, reflecting our growing research output in radiology AI.

HOW WE DIVIDE OUR WORK

Four pillars. One mission.

Each pillar reinforces the others to form the foundation for safe and responsible healthcare AI.

Data Commons

National and International scale clinical data networks

Consortia across public and private institutions
DPDP Act compliant data sharing
Networks of public and private clinics powering national-scale AI

Click for full brief

Data Commons

We are helping build consortia needed for training foundational AI models.

Indian healthcare data is fragmented across private hospitals and under-digitised public institutions, making national-scale AI research nearly impossible today. CRASH Lab is building the consortia and the trusted infrastructure that lets this data work together - consented and governed by Indian law. Our partnerships span public and private hospitals, imaging centres, and academic institutions with extensive data-sharing networks both for research and partnerships.

Click for full brief

Standards & Benchmarks

Real-world, complex clinical benchmarks for frontier medical AI systems

Hard benchmarks grounded in real clinical cases
Bias and subgroup audits
Failure-mode taxonomies for safer deployment

Click for full brief

Standards & Benchmarks

We measure what generic healthcare benchmarks miss in AI models before they are deployed.

Most AI benchmarks have drifted from clinical reality and score models on cases they have already seen, on simplified tasks far from how medicine is practiced in the real world. CRASH Lab builds rigorous, real-world and clinically complex benchmarks - co-designed with practising clinicians and stress-tested across contexts. We also map how models fail, to help the field build something better and safer.

Click for full brief

Foundational AI Models for India

Foundation models for South Asian population

Models that work for South Asian patients
Novel geometric and topological architectures
Bias mitigation in AI models

Click for full brief

Foundational AI Models for India

Foundational Models built for our patients.

Most clinical AI today is trained on Western populations and Western workflows and it does not generalise to India. CRASH Lab develops foundation models tailored to South Asian healthcare with models that understand the patients, disease prevalence and the clinical reality of the subcontinent. We pair this with novel architectures that move beyond brute-force scaling, and bias mitigation built into the model itself.

Click for full brief

Human-Centric Multi-Agent Systems

Clinician-centred AI, with framing of policies built for what comes next

Multi-agent systems co-designed with clinicians
Forward-looking policies for autonomous AI
Trusted, edge-deployable AI agents

Click for full brief

Human-Centric Multi-Agent Systems

AI co-built with clinicians for the future of care delivery.

We believe that the future of clinical AI is not a single model but a team of specialised agents working alongside clinicians at the point of care. CRASH Lab is building human-centric multi-agent systems designed with frontline clinicians, deployable at the edge and grounded in policies for the autonomous era of healthcare. Technical capability without governance is unsafe and governance without technical capability is performative theatrics and we work on both.

Click for full brief

Each pillar feeds the next: data → benchmarks → models → deployment.

Let’s Accelerate Healthcare AI Innovation Together

Whether you're a clinician, researcher, or industry partner — we'd love to collaborate.

Get involved