Reimagining Oncology Workflows: A Conversation with Talha Basit of Spiraldot Health

Talha Basit, co-founder and CEO of Spiraldot Health, is a healthcare innovator focused on harnessing artificial intelligence to transform clinical workflows. With a background spanning healthcare consulting, technology, and entrepreneurship, he brings both technical expertise and strategic insight to solving some of the most pressing inefficiencies in healthcare delivery, starting with oncology. In this interview, Talha shares insights into Spiraldot Health's unique approach to training and fine-tuning AI models, detailing their methodology built on four interconnected pillars: a clinical-first data foundation, a rapid iteration training pipeline, continuous clinical validation, and robust post-deployment monitoring systems.

Interview

Q: Spiraldot Health is taking a fresh approach to integrating AI into clinical workflows. What core problem were you trying to solve when you founded the company, and how has that focus evolved?

A: When we started, we were really focused on three core problems we kept seeing in cancer care. First, so many of the workflows were incredibly time-consuming, manual, and prone to error. Second, we wanted to dramatically save clinicians’ time on things like research and matching patients to clinical trials, because that’s time better spent with patients. And finally, we wanted to make sure that every patient, no matter their location or circumstance, had access to a treatment plan based on the absolute highest quality, most current clinical evidence.

Those goals haven't changed at all. If anything, the more we learn about the challenges in oncology, the more convinced we are that our approach is not just valuable, but necessary.

Q: Can you walk us through your approach to training and fine-tuning your AI models? What distinguishes your model architecture or data strategy from others in the healthcare AI space?

A: Our approach to training and fine-tuning AI models is built around a fundamental principle: clinical excellence at scale. Rather than developing everything from scratch, we strategically leverage foundation models while creating proprietary capabilities that directly address the complex needs of healthcare providers and their patients.

Our methodology centers on four interconnected pillars that deliver measurable clinical and business outcomes.

First is our data foundation. For us, data governance is much more than a compliance checkbox. We use multi-layered de-identification, keep a full audit trail, and are completely HIPAA compliant. Additionally, we've invested significantly in proprietary synthetic data generation capabilities, allowing us to augment real-world datasets while maintaining absolute patient privacy protection.Every data preparation decision is driven by clear clinical objectives: What specific capabilities are we enabling? How will our fine-tuned models improve patient outcomes? This clinical-first approach to data preparation ensures we're building solutions that translate directly into healthcare value.

Next is our training pipeline. We favor foundation models as our starting point, but our differentiation lies in our systematic approach to clinical specialization. Our training pipeline follows a carefully orchestrated progression designed for rapid iteration and clinical validation.We begin with pre-training on comprehensive medical literature and clinical documentation, establishing a robust foundation of medical knowledge. Our team works directly with practicing clinicians to formalize clinical workflows that traditionally exist only in the minds of experienced practitioners—essentially digitizing decades of clinical expertise into our models.Our ability to iterate quickly gives us a significant competitive advantage in responding to emerging clinical needs.

Third, and this is where I think we really stand out, is our process of continuous clinical validation. Rather than treating validation as a final checkpoint, we've embedded clinical review throughout our entire training process. We aim to have our clinical advisory network validate our model outputs at every training milestone. This ensures our models don't just excel on technical benchmarks—they align with real-world clinical decision-making processes and existing healthcare workflows.

And finally, our job isn't done once a model is deployed. We leverage and build when needed monitoring systems that track model performance across diverse real-world clinical settings, providing continuous feedback that drives iterative improvements. Healthcare providers using our solutions can contribute to this feedback ecosystem, helping us identify edge cases and optimization opportunities. This creates a virtuous cycle of improvement that benefits our entire customer base. We maintain robust safety protocols, including automatic rollback capabilities if performance metrics fall below established clinical thresholds. The old saying that “feedback is a gift”, holds true for all of us in the HealthCare space. Anyone who is not listening, or doesn’t prioritize this, won’t remain here for long.

While this is not focused on the AI model we do consider our strategic vision another thing that sets us apart. Our approach balances innovation with the safety-first culture that healthcare demands, but we're not content with incremental improvements. We're building the infrastructure for the next generation of healthcare AI—models that don't just assist clinicians, but actively improve patient outcomes at population scale. This methodology has positioned us to capture significant market share while maintaining the clinical rigor and safety standards that healthcare organizations require. As we continue scaling, every technical decision reinforces our core mission: transforming healthcare delivery through responsible AI innovation.

Q: How do you ensure that your models are clinically reliable and safe, especially as you’re working within environments with high consequences for error?

A: Every software engineer knows about "Defensive Programming"—the idea of building your code to anticipate and handle errors. A friend of mine calls the equivalent in baseball "Defensive Excellence." That’s exactly how we see clinical reliability safety. We know there's no margin for error, so we practice defensive excellence at every single level.

We don't just rely on one safety measure. We put our models through rigorous testing with datasets it has never seen before. Our goal is always human-AI collaboration. When we think about our models it is never in the mindset to replace the clinical decision, but to augment it. It’s why we include confidence intervals and uncertainty quantifications in our outputs – so clinicians know exactly how much weight to give our recommendations. We flag uncertainty rather than provide potentially unreliable outputs. 

Everyone knows in Healthcare, earning trust takes years but losing it happens in minutes. We prioritize long-term credibility over short-term performance gains, because that's what sustainable success requires.

Q: Many AI tools struggle with integration into existing clinical workflows. How does Spiraldot streamline or enhance the day-to-day tasks of frontline clinicians?

A: You've hit on a problem that is absolutely core to our solution and a huge differentiator for us. We are not a separate application with another set of logins and passwords. Instead we are directly embedded in the clinicians existing EHR systems with nothing new to learn. We feel very strongly about meeting clinicians where they are. We further streamline the existing workflow which might require clinicians to login into 6 or 7 different applications into a single, unified pane of glass. Greatly enhancing productivity for frontline clinicians. 

Q: You have clinicians on your founding team and clearly embed clinical expertise into product development. How has that shaped the direction and usability of your platform?

A: Having clinicians on our founding team has been everything. The clinicians on our team ensure that all our solutions are built with the clinicians workflow in mind. Our physicians remind us daily to take their POV and not try to change workflow, but simplify and enhance them. 

Q: There’s often concern about over-reliance on AI in diagnosis. How do you ensure your tools support, rather than replace, clinical judgment?

A: We feel very strongly about supporting rather than replacing physicians' clinical judgement. Our focus is to provide great co-pilot and assistant capabilities making sure clinicians have all the data and information they need to make the best decision for their patients. 

Q: What role does explainability play in your diagnostic models, and how do you ensure clinicians trust and understand the outputs?

A: For us, explainability is essential to how our tool works. We’re not just showing AI outputs. We’re helping oncologists make real decisions and our reasoning has to be clear. For every recommendation, we show the underlying data (labs, genomics, prior treatments) and connect it directly to clinical guidelines like NCCN. Clinicians see why the AI is suggesting something, not just what it suggests.

We also embed everything into the EMR and tumor board workflow, so it feels like part of how they already work. Our strength is that we don't alter the physician workflow. Also, we support feedback: doctors can agree, disagree, or adjust, which helps us keep improving the model in a way that stays grounded in real-world care.

As you scale, how do you see Spiral Health fitting into the broader landscape of AI-powered healthcare tools? What’s next in your roadmap that you’re especially excited about?

Our vision is to become the leading global platform for complex disease care.

As for what's next, I’m personally very excited about a new visual tool we're launching for NCCN Guidelines, which will make navigating those complex treatment pathways much more intuitive. We’re also releasing a new tool to help streamline insurance eligibility.