Loka and Biocient Build an AI Research Assistant for Faster Alzheimer's Data Analysis
Industries
Life Sciences,
Biotechnology
Tech & TOOLS
Amazon Bedrock, Amazon SageMaker, Amazon ECS, AWS Fargate, Amazon S3, Amazon ECR, Amazon Route 53, AWS Secrets Manager, Chainlit
Teams & Services
AI/ML, DevOps,
Project Management
milestones
Accelerating Alzheimer's research with AI on AWS
Industry
Life Sciences,
Biotechnology
Tech & TOOLS
Amazon Bedrock, Amazon SageMaker, Amazon ECS, AWS Fargate, Amazon S3, Amazon ECR, Amazon Route 53, AWS Secrets Manager, Chainlit
Teams & Services
AI/ML, DevOps,
Project Management
milestones
Accelerating Alzheimer's research with AI on AWS
Biocient is advancing therapies for neurodegenerative diseases, generating vast amounts of experimental data that researchers must analyze quickly and accurately.
The situation
Biocient develops therapies for neurodegenerative diseases and generates large volumes of experimental data, from behavioral studies and biochemical assays to microscopy images. Turning that data into insight meant significant manual effort across multiple, disconnected analysis tools, slowing research and putting advanced data exploration out of reach for many scientists.
The challenge
Researchers needed to move from raw experimental data to answers in minutes, not hours, without writing code or switching between tools. That meant handling two very different data types in one place: structured tabular datasets and complex microscopy images.
The solution also had to go beyond a simple query tool. It had to understand natural-language questions, run statistical analyses and generate visualizations on demand, and interpret microscopy images accurately enough to support pathology work. All of this needed to happen inside a secure, AWS-native environment that kept sensitive research data protected.
“Researchers can move from raw experimental data to actionable scientific insights in minutes rather than hours.”
The solution
Loka designed and built an autonomous, AI-powered data analysis platform on AWS that combines large language models, agentic workflows, and machine learning in a single conversational interface. Researchers upload experimental datasets, ask questions in natural language, and receive statistical analyses and visualizations in return. The same interface extends to microscopy image analysis through a custom Amazon SageMaker model.
Running on a scalable, AWS-native architecture powered by Amazon Bedrock, Amazon ECS on AWS Fargate, and Amazon S3, the platform replaced scripting and disconnected analysis tools with conversational AI, taking researchers from raw experimental data to actionable scientific insight in minutes rather than hours.
What we delivered
Conversational data analysis:Scientists analyze structured experimental data through natural language, with automated statistical analysis, code execution, and visualization generation.
Secure, AWS-native data handling: Datasets are managed securely within AWS throughout upload, analysis, and interpretation.
AI-powered image interpretation: A custom U-Net computer vision model on Amazon SageMaker handles microscopy segmentation and plaque detection, extending the assistant from tabular data to pathology images.
Against manually generated reports, plaque detection stayed within 8.7% undercounting to 13.2% overcounting, demonstrating reliable performance without systematic bias.
Project results
The assistant turned days of fragmented, manual analysis into a single conversational experience, giving Biocient researchers faster, more accessible, and more reliable insight across both tabular data and microscopy images.
< 3 min: Conversational AI cut tabular data analysis from 2 to 3 hours down to under 3 minutes.
10 to 15 sec: Automated microscopy image segmentation per image, accelerating pathology analysis workflows.
0.92 DICE: Accuracy of the custom Amazon SageMaker computer vision model across semantic segmentation tasks.
0.88 IoU: Highly accurate, AI-assisted identification of pathological features.
