Nijat Hasanli, Head of Product at Lindus Health – Interview Series

Nijat Hasanli, Head of Product at Lindus Health, brings a focused track record in building and scaling product functions across healthtech and technology-driven organizations, currently leading product strategy and execution at Lindus since 2022 after priors at OneCommerce and Downforce Technologies. His experience spans multiple product environments in the UK, where he has been responsible for aligning product development with business outcomes, driving innovation, and translating complex technical capabilities into scalable, user-focused solutions within fast-moving, industries.

Lindus Health is an AI-driven clinical trial company operating as an “Accountable Research Organization,” designed to give biotech and pharmaceutical companies greater control, speed, and reliability in running clinical studies. The company replaces traditional contract research models with a fully integrated, technology-first platform that manages everything from trial design and patient recruitment to data capture and execution within a single system, often completing trials significantly faster than industry norms. Its proprietary AI-native operating system enables real-time visibility into trial performance, aligns incentives through milestone-based pricing, and leverages large-scale health data to improve enrollment and outcomes, with the broader goal of accelerating the delivery of new treatments to patients.

Could you share a defining moment or early challenge that helped shape Lindus Health’s mission or product direction?

In clinical trials, innovation is often associated with scale: large organizations, significant capital, and established infrastructure. But our defining early moment taught us the opposite.

When we ran our first trial, we were deliberate about keeping it simple – something we knew we could execute safely with our technology. That constraint forced us to simplify the trial design and rely on systems we fully controlled, which made dependencies and inefficiencies more visible. Large, highly complex software stacks are not required to innovate in trial delivery. Innovation in trial delivery depends on ownership of the full trial, start to finish, with a coherent data pipeline connecting it all. That visibility lets you see how everything fits together, which surfaces opportunities that are difficult to identify in fragmented operating models.

How does your AI-powered platform streamline the clinical trial process compared to traditional models?

We don’t position this as an AI-powered platform. Citrus^TM is an AI-native trial operating system. We take a distributed approach to AI application – rather than betting big on AI in one domain, we tap our engineering team to find reliable, context-specific applications across our systems. What makes this work is that we run trials with full-scope execution. When AI speeds up protocol parsing, that efficiency flows into study design. When study design is faster, the data team inherits a cleaner setup. These incremental improvements compound because each stage produces more structured outputs and reduces downstream rework.

Two examples illustrate this: we use AI to generate data analysis code, where we have clear mechanisms to review the output before it touches anything. We also use AI to parse protocol documents into our study design schematics – significant efficiency gains on that first transformation, but the build still goes through weeks of review and acceptance testing.

We do not apply AI in clinical care workflows. We feel a responsibility to ensure our applications pose no risk of harm to patients in our trials because the infrastructure required to support appropriate guardrails is still maturing. This avoids the regulatory and safety risks associated with AI in clinical care workflows. Our other applications still face the same scrutiny as any AI in this industry – but fundamentally, they offer plenty of time for human review. Clinical care is a different risk profile: in-the-moment decisions where errors could directly impact patient safety.

The primary impact is on clinical research team efficiency. Much of our streamlining serves the clinical research team directly – efficiencies that compound. Every week we shave off trial delivery is a week closer to completing the drug development cycle, which can reduce overall timelines in the drug development process. And beyond timeline gains, our AI applications do more than forecast recruitment. They monitor incoming data for anomalies, safety signals, and risk indicators – giving the study team clearer visibility to oversee trials effectively.

In traditional models, AI is typically applied as a point solution – one tool for one function, isolated from the rest of the process. Our end-to-end model lets efficiency gains flow through the entire trial, with each improvement building on the last. It reflects the difference between optimizing individual functions and improving performance across the full trial execution scope.

The Tufts study highlights that nearly one-third of clinical trial data is non-essential. How is Lindus Health using AI to identify and eliminate unnecessary data collection?

What the Tufts study describes is something we’ve seen firsthand. In our experience, the root cause is structural: when trial delivery is fragmented across teams, each team is incentivized to cover all their bases. This behavior is a structural response to fragmented accountability. From vendor contracts to data collection instruments, every handoff adds another layer of precaution.

By the time you reach the person designing the case report forms, they’re often far removed from the original research question. The investigator might be asking, “How did weight change over three months on this treatment?” But the instrument designer is focused on operational concerns – site payment triggers, compliance checkboxes, audit trail requirements. Both perspectives are valid and necessary. The issue is that these functions operate without shared context. The operational scaffolding grows, and the research question gets buried.

There is also a simpler layer to address before technical solutions. Our research teams already have access to AI chat tools, and they use them constantly. When teams receive dozens of protocol documents and PDFs every week, being able to input them into AI-assisted tools and ask questions changes how you engage with that material. This helps teams stay closer to the research question instead of getting lost in operational detail.

First, trial design. Having AI tools embedded across the design and build process unlocks opportunities to catch these issues early – before they become locked into the protocol. We use AI to generate refined study schedules and guide designers through the protocol, flagging where the research plan is getting overcomplicated, where there’s duplication, or where errors have crept in. From there, the designer can make an informed decision to remove a data point or reduce collection frequency – with the reasoning documented.

Second, data analysis. Once a trial is running, changing what you planned to collect is a different challenge. But AI can help teams cut through the noise faster – quicker aggregation, pattern detection, and anomaly flagging means less time on manual processing. That matters here because it gives teams the agility to identify if and where unnecessary data collection is actually impacting the trial. With confident insights arriving sooner, they can make adjustments, flag the issue to the research team, or build the case for a protocol amendment to streamline collection while there’s still time to act.

This is a structural issue addressed at two stages: in design, to catch and remove complexity before it’s locked in, and in analysis, to give teams the speed to identify issues and act on them while the trial is running.

What are the biggest misconceptions about using AI in clinical trials, and how do you address them with sponsors and regulators?

The biggest misconception is hesitancy – the assumption that sponsors and regulators would resist AI in clinical trials. This is not reflected in our experience.

On the regulatory side, our early conversations have shown that while regulators are appropriately cautious about publishing guidance, the individuals in these organizations are open to discussing AI in clinical trials. There’s a general recognition that AI can improve productivity, efficiency, and quality – and an awareness of how much redundancy exists in this industry.

On the sponsor side, we’ve had sponsors ask us about AI before it was introduced in discussions. They were actively looking for us to find and implement these solutions. This is driven by two factors: first, sponsors are already using some of these tools themselves, so they understand the potential. Second, they recognize that AI-driven efficiencies could cut trial duration, reduce costs, and prevent issues that might otherwise have gone unnoticed. We’ve heard from several sponsors that there’s internal pressure in their organizations to demonstrate AI use for efficiency.

A related concern is about AI and data being used to train models. The industry is coming along here, and model providers are increasingly clear about how their usage plans work. We’re careful to ensure the AI tools we use are not feeding data into foundation model training. We’re equally careful about training our own models or methodologies – and where anyone does this, they should be explicit about it in the statement of work between customer and provider. Clear documentation of data usage and model behavior is required.

So the misconceptions are real, but they point to a duty of responsibility: be clear about where AI is used, how data is handled, and what safeguards are in place. The relevant question for sponsors isn’t whether to use AI in trials – it’s whether their provider has thought through these issues and is willing to be transparent about it.

How do you balance automation and human oversight to ensure both speed and quality in trial execution?

We’re cautious about using AI where there’s no opportunity for human review. This is reflected in our quality documentation and AI policy.

To illustrate: there are providers offering chatbots that assess patient eligibility through conversation. This kind of automation needs far more careful consideration than most applications. Best case, the AI wrongly disqualifies someone from a trial that could have helped them. Worst case, it qualifies them, signals to the study team that they’ve passed screening, and introduces risk into enrollment that shouldn’t have been there.

Human oversight doesn’t help here – by the time a human reviews the output, the AI has already acted in a high-stakes workflow. Compare that to a case report form error: if data collection goes wrong, you can adjust the instrument or discard the data. But if AI tells a patient or a research site to take an action, the potential for irreversible harm is greater – both in severity and compared to other places AI might go wrong in a trial.

This balance is achievable by focusing on applications where human review is integrated and risks are manageable. AI is most effective in workflows where human review is embedded and errors are recoverable.

What technologies or design principles are most effective in reducing patient burden and improving retention?

Patient burden and retention comes down to small, deliberate steps to ensure a good experience. No single intervention addresses this independently.

Two design principles matter most.

First, the quality of patient-facing content and interfaces. The patient information sheet, the consent form, the application they use during the trial – all of these shape the experience. Copy should be plain and concise. Interfaces should be simple: intuitive navigation, minimal friction, no buried documents. Patient advocacy consultations can help refine these materials before they reach participants. Good user experience design matters as much here as anywhere, particularly in clinical trials where you’re not dealing with millions of users who’ll eventually adapt.

Second, how research teams stay connected to patients. This means communication tools – reminders, invitations, notifications – and monitoring infrastructure that surfaces patient status, adherence, and safety signals. Automated oversight scripts help here, flagging what needs attention so teams can respond promptly. Machine learning can detect patterns in adherence data – early signs of disengagement before a patient drops out – enabling proactive intervention rather than reactive follow-up. The goal is the right information to the right people at the right time, without noise that dilutes what matters.

These improvements do not rely on novel technology, but they don’t come from off-the-shelf solutions either – there’s no product you can drop into a trial and expect to work. It takes attention: understanding where patients hit friction and addressing it deliberately. What modern AI tooling offers is a way to do this faster – refining copy, reviewing tone, automating monitoring scripts. The technology is mature. The difference is whether you’re solving for the trial or solving for the patient.

How does Lindus Health gather and integrate patient feedback into trial design while keeping the process lean and efficient?

Privacy and compliance requirements shape how patient feedback can be collected – the approach has to work within those boundaries. Basic monitoring captures service uptime, de-identified usage data (device type, application behavior), and adherence patterns – how consistently participants complete scheduled assessments. When this data surfaces friction points, it feeds into design decisions for future trials.

The more direct integration comes through research staff. Coordinators are encouraged to engage with participants and collect signal around their experience, then feed that back to the wider team. This is reinforced culturally – patient experience feedback gets published in shared channels and called out at company all-hands.

There’s also a structural advantage. Unlike the traditional model, where a new research team is assembled for every study, Lindus runs trials on the same technology with team members who have worked across multiple studies. This continuity allows learnings – both codified and tacit – to flow from one trial into the design of the next. When a coordinator encounters friction in one study, that insight can inform how the next study is set up.

Patient advocacy groups extend this further, surfacing perspectives that wouldn’t come through internal channels – particularly around how study materials and processes land with different patient populations.

The process remains lean because feedback flows through existing structures rather than requiring a separate apparatus for each study.

What needs to change industry-wide for clinical research to become faster and more reliable?

The industry has structural inertia that requires practitioners to operate differently and demonstrate alternative approaches in practice. Corporate innovation programs and executive mandates have limited impact on operational change – what’s needed is practitioners who will actually do things differently and prove it works.

Statistical programming illustrates the pattern. This is skilled work – transforming clinical data for biostatistical analysis – performed by specialists who have deep domain expertise. But the struggles to attract talent. Professionals with data science or engineering backgrounds rarely choose it, even though the skillsets overlap significantly. The work remains siloed, the methods opaque to outsiders, and the talent pipeline constrained.

AI could open this up – modern tools can handle much of the transformation work, and double programming requirements (where two programmers independently produce outputs) can be met with AI-human pairs rather than human-human pairs. But technology alone doesn’t solve structural problems. You need practitioners willing to implement it thoughtfully and prove it meets regulatory standards. Without that, the capability remains underutilized.

The broader lesson: faster, more reliable trials require more than new tools. They require creating space – in hiring, in regulatory interpretation, in organizational culture – for people who will work differently. This is how timelines can be reduced in practice.

How do you see the relationship between AI, data, and trial design evolving over the next five years?

The relationship will be shaped by a structural reality: AI’s effectiveness is limited by the quality of context it receives. Without rich, accurate context – where data came from, what transformations it underwent, what it actually means – even strong models produce unreliable outputs.

Most of the clinical trials industry is fragmented. CROs see fragments of the trial lifecycle. Sponsors work with multiple vendors, each holding a piece of the picture. Context is lost at every handoff. When you ask an AI system to reason about trial data in this environment, it’s working with incomplete information – and incomplete information produces unreliable outputs.

The organizations that will benefit most from AI are those with end-to-end traceability. They control the data chain from protocol design to data capture to analysis. They don’t infer context – they generate it. They design the forms, define the fields, write the data dictionary. Every data point has provenance because the organization created the provenance. That traceability isn’t just operational efficiency – it’s what enables confident decisions that affect patients.

Over the next five years, this structural advantage will compound. Organizations with end-to-end visibility will deploy AI more effectively – from adaptive trial designs that respond to incoming data, to protocol optimizations informed by historical patterns – capture learnings that improve their systems, and widen the gap. Those working with fragmented data will find AI promising but unreliable: systems that perform well in controlled settings but do not generalize reliably in production environments.

The question for the industry isn’t whether AI will matter. It’s whether the data infrastructure exists to make AI trustworthy. For most of the industry, it doesn’t yet. That’s the work ahead.

Thank you for the great interview, readers who wish to learn more about AI-native clinical trial execution, end-to-end data ownership, and faster study delivery should visit Lindus Health.