FabAgent: An Agentic Optimization Framework for Sustainable Fabrics
Using Large Language Models and Evolutionary Algorithms to design sustainable, high-performing fabric blends
Anusha Narayan
The Nueva School, San Mateo, CA
About
Our Story
FabAgent began as a passion project driven by a simple question: Can artificial intelligence help make fashion more sustainable?
As a student at The Nueva School in San Mateo, California, I became increasingly aware of the fashion industry’s massive environmental footprint. With the industry responsible for 4 billion tons of COâ‚‚ emissions annually, I knew there had to be a better way to design fabrics that balance sustainability with performance.
After months of research, development, and countless iterations, FabAgent emerged — an AI-powered framework that combines Large Language Models with evolutionary algorithms to discover optimal fabric blends. The system autonomously collects data from over 2,000 scientific sources, evaluates 101 different materials across 24 metrics, and generates Pareto-optimal solutions that outperform commercial alternatives.
This project represents my commitment to using technology for environmental good, demonstrating that sustainability and performance don't have to be trade-offs.
Presenting FabAgent at JSHS 2025
Recognition
Awards & Honors
Regeneron ISEF Finalist
International Science and Engineering Fair 2025
Category: Environmental Engineering
AAAI Special Award
Association for the Advancement of Artificial Intelligence
ISEF Special Award
ACSEF Grand Prize Winner
Alameda County Science & Engineering Fair
Grand Prize
JSHS Regional Winner
Junior Science & Humanities Symposium Northern CA 2025
2nd Prize Environmental Enginnering
ICLR 2025
Workshop paper in Tackling Climate Change with Machine Learning Track
Peer Reviewed
​
US Patent Pending
SYSTEMS AND METHODS FOR AN AGENTIC OPTIMIZATION FRAMEWORK FOR SUSTAINABLE FABRIC BLENDS
Serial No.: 19/056,486
The Problem
Fashion’s Hidden Environmental Crisis
The fashion industry emits 4 billion tons of COâ‚‚ annually, contributing 6-7% of global greenhouse gas emissions and more than aviation and maritime shipping combined.
​
A full one-third of the fashion industry's emissions stem directly from the choice of fabric blends used in textiles. For example, producing just 1kg of cotton emits 3.3 kg of COâ‚‚, while 1 kg of polyester emits 20 kg of COâ‚‚.
​
By 2050, the fashion industry could account for 26% of the worldwide carbon budget associated with the 2°C warming limit.
Current Approaches Fall Short
​
-
Datasets on fiber properties are limited and not easily accessible
-
Studies using GA and ANN are limited to yarn properties
-
Previous experiments limited to 2-component blends
-
No comprehensive multi-objective optimization framework
The Textile Waste Crisis
The Solution
FabAgent: AI-Optimized Fabric Design
Using Large Language Models and Evolutionary Algorithms to optimize fabric blends that minimize environmental impact while maximizing durability and comfort.
Methodology
AI-Driven Fabric Optimization
System Overview
AI-Driven Fabric Optimization
Data Collection Architecture
FabAgent iterates over every material-metric pair, autonomously searching and aggregating data from web sources and scientific papers using the ReAct framework.
101 Raw Materials
24 Metrics
2000+ Sources
Action-Oriented Query Generation
The Search Function converts natural language prompts into structured search actions, executed via SerpAPI to access Google results programmatically.
LangChain
SerpAPI
ReAct Framework
Observation & Reasoning
The agent analyzes search results, generates observations, and determines whether to make follow-up searches or output final JSON with values and source citations.
GPT-4
JSON Output
Source Citations
Evolutionary Optimization
Implements NSGA-II (Non-Dominated Sorting Genetic Algorithm II) in Pymoo to find Pareto-optimal blends that balance sustainability, cost, comfort, and durability.
NSGA-II
Pymoo
Pareto Front
KNN Imputation
24 Material Metrics
FabAgent evaluates each material across 24 carefully selected metrics spanning four key categories. Environment and cost are minimized, while durability and comfort are maximized.
Category / Metrics (6 per category)
Environmental
Minimize ↓
Water Consumption (L/kg) · GHG Emissions (kg COâ‚‚e/kg) · Land Use (m²/kg) · Biodegradation Time · Energy Consumption (MJ/kg) · Recyclability Score
Durability
Maximize ↑
Tensile Strength (MPa) · Abrasion Resistance · UV Resistance · Pilling Resistance · Tear Strength · Wash Durability
Comfort
Maximize ↑
Moisture Regain (%) · Air Permeability · Thermal Conductivity · Softness/Hand Feel · Drape Coefficient · Elasticity/Stretch
Cost
Minimize ↓
Raw Material Cost ($/kg) · Processing Cost ($/kg) · Transportation Cost · Storage Requirements · Availability Index · Market Volatility
Agent Pipeline
How FabAgent Works
Results
Outperforming Industry Leaders
FabAgent vs. Commercial Blends
Benchmarked against leading fashion brands: GAP, Giorgio Armani, Nike, and Banana Republic.
↓ Lower is better for Environment and Cost. ↑ Higher is better for Durability and Comfort.
Maximum Sustainability Gain
+62%
Improvement over commercial blends with K=3 configuration
Comfort Improvement
+91%
Enhanced comfort metrics while maintaining durability
vs. Nike Shorts
+44%
Sustainability improvement while beating cost, durability, and comfort
Discovered Blends
Optimal Fabric Compositions
Ongoing Research
Continuation Work
Industry Validation Survey
Conducted comprehensive industry validation through an expert panel survey (N=40) with textile and fashion professionals. 82.1% of experts indicate high likelihood to adopt FabAgent, and 92.3% view it as an improvement over current processes.
Expert Panel
Industry Feedback
Validation Study
Physical Validation
Producing FabAgent-generated blends in collaboration with laboratories to validate FabAgent's framework.
NC State University
Laboratory Testing
Material Synthesis
Research in progress — publication forthcoming.
Conclusion
A New Paradigm for Sustainable Fashion
-
Created one of the largest datasets in the field: 101 fibers across 24 attributes mined from 2000+ web sources and scientific literature
-
Demonstrated that AI-driven optimization can discover fabric blends that outperform commercially available blends from leading fashion brands
-
Developed a highly scalable approach for autonomous data collection that can be applied to other fields
-
Proved that sustainability and performance don't have to be trade-offs: optimized blends can excel in both
What’s Next
Future Applications
Sustainable fabric was just the beginning. FabAgent is the first generative AI framework for material blend optimization—a platform designed to revolutionize material discovery across industries.
Battery Materials
Pharmaceuticals
Construction Materials
Bioplastics