Healthcare’s AI Test

Superadmin

Apr 1, 2026 - 11:02

0 2

Healthcare AI is almost always framed as a promise — sharper diagnostics, faster decisions, better outcomes.

In India, that promise is now beginning to take shape across radiology labs, pathology workflows, and public health programmes.

In a country that severely lags in access to healthcare, the doctor-population ratio in India for the allopathic treatment category is estimated to be 1:1200 (as of late 2025), which is below the WHO recommended level of 1:1000. And while that disparity does not seem grave, this is based on 80% availability of approximately 13.88 Lakh registered allopathic doctors for a population of over 1.4 Bn.

While we are seeing plenty of healthcare AI startups, the tension between the promise and on-ground reality is something that is yet to be resolved. Can AI truly solve this gap? Because if so, no other country needs this to happen as much as India.

From Models To Medical Infrastructure

The first wave of healthcare AI was about building models — improving accuracy, training on more data, and proving technical feasibility. That phase has matured.

Ankit Modi, cofounder, and chief strategy and growth officer at Qure.ai, which makes AI for medical imaging, says the real gap today lies in deployment.

In many hospitals, AI still operates as an add-on. It analyses data but doesn’t actively participate in clinical workflows. The shift now is toward making AI a native layer within healthcare systems — embedded across screening, diagnosis, and follow-ups.

Jadeja Dushyantsinh Anopsinh, AI advisor for eye screening startup Remidio, adds that much of healthcare AI is still focused on the wrong layer. “The real whitespace is at the risk triaging layer, before patients ever reach a specialist,” he says.

In one of Remidio’s programmes in Kerala, for instance, nearly 99% of diabetic retinopathy cases detected were previously unknown to patients, highlighting how hospital-centric AI misses large parts of the population that never enter formal care systems.

“There’s also a deeper imbalance… much of AI innovation is concentrated in high-resource environments, however, the biggest challenges… start at the first point of care,” Modi added.

The focus is shifting from intelligence to integration.

Cracking Clinical AI And Real-World Use

Then comes the clinical layer, where the potential for AI applications is hardest to determine. This is because most medical and clinical treatments are designed for administration by healthcare professionals. This decision-making cannot simply be passed on to AI systems.

This is why many like Kalyan Sivasailam, the cofounder of radiology platform 5C Network, call it the most complex problem to solve in healthcare AI.

Clinical systems need to be consistently reliable in high-stakes environments. “The challenge, even today, is just getting good enough. Good enough to be able to consistently rely on something in the midst of clinical practice,” the 5C Network founder told Inc42.

To address this, companies are moving toward agentic AI — systems that don’t just produce a single output but reason through multiple steps, validate findings, and mimic how clinicians arrive at decisions.

At the same time, Dr. Amit Kharat, cofounder of DeepTek AI, an AI radiology startup, highlights that radiology remains one of the most active areas for AI deployment. The combination of structured imaging data and a shortage of radiologists has made it a natural entry point.

DeepTek’s approach combines platform infrastructure, multiple AI integrations, and reporting workflows. Hospitals can triage scans, prioritise critical cases, and generate draft reports — reducing turnaround time and easing clinician workload.

On the ground, AI is already showing impact in specific use cases.

At Qure.ai, routine diagnostics are being repurposed for early detection. In Punjab, AI integrated into stroke workflows has reduced turnaround time by 85%, enabling faster clinical decisions.

AI As Co-Pilot, Not Replacement

Despite growing adoption, hesitation persists — largely driven by the perception that AI could replace clinicians.

Tathagato Rai Dastidar, founder of SigTuple, a company which enabled AI-powered pathology microscopy, counters this directly. “Today, it is not replacing the pathologist — but it is fundamentally transforming how microscopic review is done.”

AI is already handling large parts of the diagnostic workflow, pre-classifying samples, flagging abnormalities, and identifying rare cases, allowing clinicians to focus on complex decisions. Other founders say AI now supports doctors across the entire care journey — from symptom triaging and drafting prescriptions to follow-up communication — while keeping the clinician in control.

Adding to this, Arjun Nagulapally, CTO at AIONOS, points to AI agents taking over coordination-heavy layers in healthcare — from discharge planning and insurance workflows to patient communication. These systems can surface relevant insights during consultations, automate administrative overhead, and ensure patients leave with clear, actionable instructions.

This is particularly relevant in India, where access gaps are significant. Platforms built around AI and digitisation are enabling remote diagnostics, connecting smaller centres with specialist expertise.

Imaging data can be analysed remotely, enabling faster reporting and expanding access to quality care.

Across implementations, AI is emerging as a co-pilot — improving consistency, reducing turnaround time, and extending the reach of clinicians.

What’s Slowing Adoption And What Comes Next

Even as use cases expand, scaling AI across India’s healthcare system remains uneven.

Dastidar points to three recurring barriers: fear of redundancy, lack of incentives for quality, and unrealistic expectations of AI replacing clinicians entirely.

From a broader system perspective, Ajay Mahipal, cofounder and general partner at HealthKois noted that while AI solutions are working, widespread deployment is struggling at the moment. Fragmented health data systems, limited interoperability, and unclear data governance frameworks continue to slow integration.

Kharat adds that deployment itself is complex. “Deploying AI [in healthcare] is not plug-and-play… there has to be sensitisation, training, and workflow integration,” he says.

Varun Dubey, founder and CEO of Superhealth, an AI-native hospital operator, adds that adoption ultimately comes down to whether technology reduces friction for clinicians.

He argues that much of the current healthcare tech stack adds complexity rather than simplifying workflows, which is why adoption remains slow.

Adoption patterns also vary. Governments are emerging as early adopters in public health programmes like tuberculosis screening, while private adoption is being driven by teleradiology firms and diagnostic chains.

Adding to this, Dushyantsinh, notes that traditional enterprise SaaS is rarely the dominant model in healthcare.

Instead, deployments are often program-led, where governments, insurers, or health systems adopt AI based on outcomes rather than licenses.

Across markets, the commercial model varies — from reimbursement-linked public health programmes in India to value-based care models driven by insurers in the US — but the common thread is that adoption is tied to measurable impact rather than software usage.

From a venture perspective, Surya Mantha, managing partner at Capria Ventures, notes that early-stage healthcare AI remains crowded, with the key differentiator being real-world usage.

Systems that operate in live clinical environments, solve high-frequency problems, and continuously improve through proprietary data are emerging as more credible, compared to solutions that remain confined to pilots or controlled settings.

Looking ahead, several underexplored areas are emerging.

Mahipal highlights early screening in tier 3 and tier 4 regions as a major opportunity, driven by high disease burden and specialist shortages. He also points to hospital operations — including inventory management and revenue cycle optimisation — as areas with significant potential.

On the clinical side, Sivasailam is focused on precision oncology and personalised medicine, while Dastidar emphasises expanding into rare and long-tail diagnostic conditions where data scarcity remains a challenge.

The Weekly Buzz: Freshworks Cofounder Says ‘SaaSpocalypse’ Won’t Happen

“AI agents won’t kill the industry.”

That’s Freshworks cofounder Girish Mathrubootham’s take as per a recent ET interview, and he’s pushing back against one of the loudest narratives to hit enterprise tech in recent memory.

The trigger was Anthropic’s Claude Cowork. On January 30, the AI startup introduced 11 plugins to its workplace automation tool — one that autonomously executes multi-step tasks directly on a user’s computer. The market reaction was swift and brutal. Nearly $285 Bn in combined market cap across SaaS, legal tech, and financial services was wiped out in a single trading session on February 4. Jefferies’ equity strategist Jeffrey Favuzza dubbed it a “SaaSPocalypse.” The Nifty IT index fell nearly 6% — its worst single-day drop in six years.

Mathrubootham isn’t buying the apocalyptic framing. His argument sits closer to Gartner analyst Anushree Verma’s read, that this is a directional shift in how services are priced and delivered, not an industry extinction event. NASSCOM has made similar noises, pointing out that real business value from AI still needs humans in the loop.

The harder-to-dismiss data point, though, is hiring. India’s top five IT companies added just 17 employees net in the first nine months of 2025-2026 financial year, reported Times of India. Growth has stalled. Recruitment is largely replacement-only now, not expansion.

SaaSPocalypse or not — the volume-led era of IT services may look like it’s behind us.

Startup In The Spotlight: Reo.Dev

Infrastructure software which includes APIs, data infra, AI tooling — is a market larger than application software. Yet every major GTM tool, from Sixth Sense to Apollo, was built for business buying journeys. But it looks nothing like this for developers.

Bengaluru and US-based Reo.Dev, founded by Achintya Gupta, Piyush Agarwal, and Gaurav Jain in 2023, was built around this gap — a developer-native GTM intelligence platform that tells infrastructure and dev tool companies exactly who is evaluating their product, before a single sales conversation happens.

Most GTM tools are built for business buyers. When developers are the ones evaluating your product, the entire playbook breaks down. The core insight of the startup is that whenever software purchase journeys change structurally, a new stack has to emerge to serve them.

Backed by investors like Heavybit and India Quotient, Reo.Dev has raised a total of $6.2 Mn thus far, and currently, it works with customers including Nvidia, LangChain, and Temporal.

The platform’s core is a de-anonymisation algorithm that surfaces which developer accounts are actively evaluating a product — across open source, GitHub activity, Docker installations, and documentation signals without those companies knowing.

Using shared GitHub repositories, Reo.Dev can display the intent behind projects, the people behind them, their reporting managers, and their technology stacks.

Prompt Of The Week

What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work?

Here’s Raunak Bhandari, cofounder & CTO of content marketing platform KiwiQ AI, with his suggestion for turning dense technical research into actionable design documents:

“You are my senior staff engineer and technical research advisor with deep expertise in ML systems and distributed infrastructure.

Here is a technical paper / RFC / research document: [paste or attach document].

And here is the system context: [paste: current architecture, scale constraints, team size, production SLAs, timeline].

– Distill the core technical contribution and key insight in 3–4 sentences a non-ML engineer could understand.

– Identify 2–3 ideas directly applicable to our system, with a frank assessment of integration complexity (trivial / moderate / significant re-architecture).

– Flag hidden assumptions in the paper that would break at our scale or in our production environment.

– Draft a one-page design doc outline covering: problem statement, proposed approach, alternatives considered with trade-offs, migration path from current state, rollback strategy, and success metrics.

Respond as a structured design doc brief I can circulate to my engineering team for an architecture review.”

Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.

[Edited by: Nikhil Subramaniam]
[Creatives by: Abhyam Gusai]

The post Healthcare’s AI Test appeared first on Inc42 Media.