Loon Lens™: Autonomous AI Agents for Literature Screening in Systematic Reviews

Addressing the Challenges of Systematic Reviews

Systematic reviews are the gold standard for evidence synthesis in healthcare, forming the foundation of clinical guidelines, regulatory decisions, and health technology assessments. However, the traditional manual process is incredibly time-consuming and resource-intensive, often taking 8-24 months to complete and costing upwards of $100,000 per review.

The exponential growth of medical literature compounds this challenge. With over 2 million new biomedical publications added annually, comprehensive literature searches now routinely yield tens of thousands of citations. Manual screening at this scale is simply inefficient. It's becoming practically impossible for research teams to manage while maintaining quality and timeliness.

The Burden of Title and Abstract Screening

The Most Labor-Intensive Phase

Title and abstract (TiAb) screening represents the most time-consuming phase of systematic reviews, often accounting for 40-60% of the total project timeline. Reviewers must evaluate thousands of citations against complex inclusion and exclusion criteria, a process that requires sustained concentration and expertise.

Human Limitations and Inconsistencies

Manual screening is inherently prone to human error and inconsistency. Studies show inter-reviewer agreement rates typically range from 70-85%, meaning different reviewers frequently disagree on whether studies should be included. Reviewer fatigue, cognitive biases, and varying interpretations of inclusion criteria all contribute to these inconsistencies, potentially compromising the quality and reproducibility of systematic reviews.

"The traditional approach to systematic reviews is unsustainable. We need innovative solutions that maintain scientific rigor while dramatically improving efficiency." — Dr. Ghayath Janoudi, CEO, Loon

Introducing Loon Lens™

Autonomous AI for Systematic Reviews

Loon Lens™ represents a breakthrough in AI-powered literature screening. Unlike traditional tools that require extensive training data or manual calibration, Loon Lens operates autonomously from day one. Simply provide your inclusion and exclusion criteria, and the system handles the rest — no coding, no complex setup, no lengthy training periods.

Key Features and Capabilities

• Explainable, HTA-grade: Outputs are traceable, transparent, and compliant with HTA requirements

• Scientific Validation: Demonstrated unparalleled performance in published benchmark studies

• Calibrated Confidence Scores: Confidence-routed human validation for unmatched 99% sensitivity and 90% precision

• Fully Autonomous: Operates without requiring pre-labeled training data

• No Pre-training Required: Works from day one without manual calibration

• User-Friendly Interface: Designed for researchers without technical AI expertise

• Scalable Performance: Handles thousands to millions of citations with consistent accuracy

• Transparent Reasoning: Provides clear explanations for every screening decision

"Loon Lens™ transforms systematic reviews from a burden into a competitive advantage. What used to take teams months or years now takes days or weeks, with better consistency and documentation." - Dr. Ghayath Janoudi, CEO, Loon

The Validation Study: Rigorous Testing Across Disciplines

Comprehensive Validation Methodology

To validate Loon Lens™'s performance, we conducted an extensive study across eight systematic reviews spanning diverse therapeutic areas. The validation encompassed 3,796 citations that had been previously screened by expert human reviewers, providing a robust benchmark for assessing AI performance. The study is available on medRxiv for full transparency.

Study Design and Scope

The validation study included systematic reviews from oncology, cardiology, infectious diseases, neurology, and rare diseases. This diversity ensured that Loon Lens™ was tested across varying levels of complexity, terminology, and study designs. Each review had been completed using traditional manual methods, providing gold-standard comparisons.

Validation Study Overview

Study Characteristics

• 8 systematic reviews analyzed

• 3,796 total citations screened

• 5 therapeutic areas covered

• Gold standard: Expert human screening

Performance Metrics

• Sensitivity (Recall): 99%

• Accuracy: 96%

• Specificity: 95%

• Precision: 90% (with 5% confidence-routed validation)

• F1 Score: 0.770

Results: Exceptional Performance Across All Metrics

Validated Performance Metrics

The validation study demonstrated Loon Lens™'s exceptional performance across all key metrics:

• 99% Sensitivity (Recall): Missing fewer than 2% of relevant studies

• 96% Accuracy: Overall correct classification rate

• 95% Specificity: Accurately excluding irrelevant studies

• 90% Precision (Peer-reviewed): High proportion of true positives among identified studies

• 0.770 F1 Score: Unmatched balance between 99% recall and 90% precision

Unprecedented Efficiency

Beyond accuracy, Loon Lens delivers transformative efficiency gains. Reviews that traditionally require 8-24 months are completed in 2-4 weeks.

Traditional vs. Loon Lens™ Systematic Review Comparison

This is how traditional and Loon AI costs and timelines compare for a Systematic Literature Review  prepared for an HTA submission.

Traditional Approach

• Timeline: 9 months

• Cost: $170,000 - $240,000

• Team: 6 reviewers

• Studies screened: 28,000

Loon Lens™ Approach

• Timeline: 3-4 weeks

• Cost: 50 - 70%

• Team: 1 reviewer (validation)

• Studies screened: 28,000

Technical Innovations Behind Loon Lens™

Advanced Agentic Systems Powered by Frontier Large Language Models

Each Loon Lens™ agent is powered by state-of-the-art foundational models built on the transformer architecture and specifically focused and orchestrated for biomedical literature. Unlike generic AI chatbots, our agentic systems specialize in and are validated to perform evidence synthesis tasks across multiple clinical domains and study desings. This domain-specific focus enables highly accurate interpretation of methods sections, results tables, and supplementary materials.

Calibrated Confidence-scored AI Outputs

Loon Lens™ has been proven to accurately identify when outputs are associated with a degree of uncertainty that may require a human intervention. This unique feature signficanly de-risks the implementation of AI systems in high stakes environments. It also allows for a much more streamlined expert-in-the-loop validation. With the use of our confiedence-based validation, an expert would only need to validate no more than 5% if the AI output.

Patent-pending Cognitive Ensemble AI Systems™

Every screening decision made by Loon Lens™ is processed by our proprietary architecture featuring an orchestrated array of 300+ specialized AI agents. Each agent is designed for a distinct task (e.g., citation screening, data extraction, etc.). Unlike monolithic LLMs, this modular agentic system achieves superior accuracy, explainability, and calibration across complex clinical domains, disease areas, and study methodologies.

"The explainability of Loon Lens™ is crucial in an HTA submission. Loon Lens™ demonstrates to HTA bodies and regulators exactly how decisions were made, with full audit trails. This transparency strengthens every submission." - Dr. Ghayath Janoudi, CEO, Loon

Seamless Implementation and Integration

Workflow Integration

Loon Lens™ is designed to enhance, not replace, your existing systematic review workflow. The platform works seamlessly with exports from popular reference management tools like Zotero and EndNote. You can continue using your preferred reference management and review coordination tools while leveraging Loon Lens™ for accelerated screening and data extraction.

Flexible Deployment Options

We offer multiple deployment models to meet different organizational needs:

• Cloud-based SaaS: Quick setup with no infrastructure requirements

• Private cloud: Dedicated instances for maximum AI speed and security

• On-premise: Full control for organizations with strict data residency requirements

Rapid Onboarding

Most teams are productive with Loon Lens™ within days. Our onboarding process includes personalized training for your specific use cases, and ongoing support. The intuitive interface means reviewers can focus on their expertise rather than learning complex software.

The Future of Evidence Synthesis

Living Systematic Reviews

Loon Lens™ enables true living systematic reviews that automatically update as new evidence emerges. Our agents continuously monitor publication databases, preprint servers, and clinical trial registries. When relevant new studies appear, they're automatically screened and flagged for inclusion, keeping your evidence base current without manual effort.

Expanding Capabilities

We're continuously expanding Loon Lens capabilities based on user needs and technological advances. Upcoming features include:

• Network meta-analysis automation

• Real-world evidence integration

• Multi-language screening capabilities

• Automated GRADE assessment

• Direct HTA dossier generation

Getting Started with Loon Lens™

Ready to transform your systematic review process? Here's how to begin:

• Schedule a personalized demo to see Loon Lens™ in action with your specific use cases

• Start with a pilot project to experience the efficiency gains firsthand

• Work with our team to optimize agents for your therapeutic areas

• Scale across your organization with our enterprise deployment options

• Join our user community to share best practices and influence product development