As artificial intelligence accelerates toward artificial general intelligence (AGI) and artificial superintelligence (ASI), an Indian startup is charting a different course—one that seeks to deeply understand human behaviour and biology, rather than merely predicting patterns from digital footprints.
Bengaluru-based Assessli has developed what it calls the world’s first Large Behavioural Model (LBM), a self-supervised AI system trained on human metadata that integrates genomics, neuropsychology, medical data, physiological signals and real-time behaviour to offer unparalleled personalisation.
“Current AI models don’t understand your adaptive data points, human behaviour. They don’t have a biological context like your medical history, genomics, mood, mind, neuropsychology,” said Suraj Biswas, Assessli’s founder and CEO, who is also a genomics and behavioural science expert. “Without mapping adaptive behaviour, AI remains generic, failing to evolve with individuals in real time.”
Assessli’s LBM is built on proprietary data collected from multiple sources: wearable devices, apps like Apple Health and Google Fit, environmental sensors, and biological inputs such as saliva and swab samples. “Every individual creates data points like this… There is no mechanism to capture this,” the founder explained. The model continuously assesses behaviour—from how users scroll on their phones to how they respond emotionally—to adapt and offer tailored advice.
The company’s patented methodology is already in use across sectors like education, healthcare, and financial services. With more than 150,000 active users and 20 million proprietary data points collected, Assessli is on track to gather over one trillion data points in the next two to three years. These will be used to dynamically benchmark users, continuously adjusting to their evolving biological and behavioural profiles.
Its approach distinguishes Assessli from competitors by integrating a complete human metadata layer, rather than relying solely on physical data or medical histories. While companies like Boston Dynamics work on training robots’ movements, and others like Liro focus only on medical data, Assessli’s model synthesises signals across domains, enabling applications from personalised learning environments to medical interventions.
Personalisation Beyond Digital Footprints
“What is meant by 95% or 99% personalisation?” Biswas asked rhetorically. “Every user has a different root cause for asking the same question because of different experiences. It should be hyper-personalisedso every individual gets a unique solution.”
Traditional AI recommendation engines operate on shallow data sets—clicks, searches, and keywords—with personalisation precision around 57%, according to a 2024 meta-analysis by Springer. Assessli’s LBM claims to erase this gap by integrating a far wider range of behavioural signals, pushing personalisation accuracy to over 99%.
“For example, in education, we are automating tasks like question generation and evaluation based on behavioural profiles,” the founder said. Teachers can group students according to strengths and weaknesses, enabling tailored learning pathways even in low-resource environments.
Real-World Applications and Tangible Results
A scenario in Assessli’s documentation shows how Aria, a 31-year-old professional, transformed her routine with LBM’s micro-interventions. By analysing her sleep patterns, stress signals, and even grocery choices, LBM helped reduce her migraines by 67%, improved savings, and guided career advancements without requiring extensive manual data entry.
In healthcare, LBM can personalise medication schedules based on genetic markers like BRCA1, optimising treatment for cancer patients. For older adults, AI companions powered by LBM provide emotional support by adapting to memory rhythms and mood fluctuations. The technology also enables “Digital Twin OS”—simulated models of individuals that help gig workers and students optimise daily schedules and workload.
Technical Innovation and Privacy-Centric Architecture
Unlike large language models (LLMs) that require thousands of GPUs to process massive image or video datasets, Assessli’s LBM handles data locally using encrypted, edge-computed signals. “We don’t expose raw biological data to GPUs,” Biswas said. By synthesising biological signals into smaller, context-rich datasets, the model operates with just 30–40 GPUs, making it 90% more cost-efficient than conventional AI.
Every 24 hours, LBM refines its global model using federated sketch learning, ensuring user privacy by only transmitting anonymised gradients rather than personal data. “DNA stays local… differential privacy adds noise to population stats… every query is mapped to a consent bit,” the company’s documentation assures.
Building a Foundation Model for the Future
Assessli’s approach is not merely about building an app. It’s about creating a global platform. “Our motto is not to create use cases. Our motto is to integrate with players, collect data, train the model, and then launch it for everyone,” Biswas emphasised.
The model is already integrated with educational platforms like iNurture and government initiatives in West Bengal and Maharashtra.
Assessli’s GTM strategy centres on forming partnerships with existing B2B players across sectors like education, healthcare, HR tech, and fintech. Rather than creating standalone products, the company integrates its personalisation model into platforms already serving millions of users, beginning with schools and universities, where automated assessments and tailored learning pathways deliver immediate benefits.
With expansion efforts in the UK and US, Assessli is positioning itself to collect diverse datasets that will enhance its model’s accuracy while gradually preparing for broader B2C applications.
In a competitive analysis, Assessli’s personalisation accuracy was compared with existing AI models: “Traditional AI guesses based on what you did. LBM knows based on who you are,” the company states. It envisions applications spanning healthcare, e-commerce, robotics, education, and more.
Investor Confidence
Assessli’s vision has attracted significant investor support, helping it scale its AI-driven personalisation model across education, healthcare, and other sectors. The company raised ₹10 lakh from the Indian Statistical Institute (ISI) Kolkata, which validated its foundational methodology early on. Building on that momentum, it secured a pre-seed round of ₹2.3 crore from angel investors, including doctors, researchers, architects, and NRIs.
Assessli successfully closed a seed round of $5 million, drawing support from a wide array of investors who believe in its long-term potential to revolutionise personalised AI. Among its funding efforts, the company has also raised $1.5 million specifically to establish its own data centre, reinforcing its commitment to building secure, privacy-conscious infrastructure.
With plans to onboard over a million users by 2027 and scale its LBM into a global foundation model, Assessli’s ambition reflects a profound shift in how AI interfaces with human experience. It’s not about answering questions—it’s about understanding the person asking them.
The post Assessli’s AI-led Behavioural Model Could Eclipse Language Models appeared first on Analytics India Magazine.
