CASPAR – AI-Assisted
Claims Substantiation Platform
A platform designed to simplify the claims substantiation process by combining AI-driven article screening, benefit extraction, and human validation into one structured workflow.
The Challenge
Claims substantiation is a complex and time-consuming process that involves searching across multiple scientific databases, reviewing large volumes of articles, identifying relevant benefits, and compiling structured reports. The process is largely manual, making it slow, inconsistent, and heavily dependent on individual expertise.
The Solution
To address this, an AI-assisted platform was designed to automate article screening, extract and organize benefits, and guide users through a structured workflow — while still allowing human validation at key decision points.
Understanding the User
- Gained a deep understanding of user workflows and challenges in the claims substantiation process
- Translated complex business and AI workflows into intuitive experiences
- Structured user journeys to simplify decision-making in a data-heavy environment
- Collaborated closely with business and technical teams to ensure feasibility and clarity
Workflow Overview
The platform follows a structured workflow that combines AI-driven processing with human validation:
- Articles are retrieved based on selected ingredients and data sources
- AI models classify, extract, and cluster relevant information
- Users review and select benefits of interest
- Additional inputs are defined to refine the scope
- Supporting evidence is selected
- A structured claims substantiation report is generated
User Workflow
1. Data Input
Users start by selecting the ingredient type and data sources to initiate the process. Simplified inputs help users quickly begin without confusion.
2. AI Processing
The system analyzes articles using classification, benefit extraction, and clustering models. It automates large-scale data analysis while maintaining structure.
3. Screening Summary
Users are presented with a summary of total and relevant articles after processing. This provides clarity on how data is filtered and refined.
4. Benefit Selection
AI-generated benefits are displayed with confidence scores, allowing users to review and select relevant ones. This supports informed decision-making with human validation.
5. Cost & Evidence Selection
Users select supporting articles, with clear visibility of paid and free sources along with total cost. This ensures transparency and avoids unexpected decisions.
7. Final Review & Output
All selected data is summarized, and a claims substantiation report is generated. This enables a seamless transition from analysis to output.
Solution
The solution was designed as a structured, user-friendly platform that simplifies a complex, AI-driven workflow. Users can initiate the process by defining inputs, after which the system leverages AI models to analyze and organize large volumes of data. The platform then guides users through reviewing insights, selecting relevant benefits, and refining inputs with clear visibility into supporting evidence and associated costs. Each stage is designed to balance automation with human validation, ensuring both efficiency and control. The workflow concludes with a consolidated view and the generation of a structured claims substantiation report.
Impact
The solution streamlined a complex research-driven workflow by reducing manual effort and introducing a more structured approach to claims substantiation.
The integration of AI-assisted processing with human validation improved clarity, consistency, and overall efficiency of the process.
Key Learnings
The solution streamlined a complex research-driven workflow by reducing manual effort and introducing a more structured approach to claims substantiation.
The integration of AI-assisted processing with human validation improved clarity, consistency, and overall efficiency of the process.
This project helped me evolve from designing interfaces to designing systems that integrate AI, user behavior, and business workflows.