Firefly-Geni Autonomous Framework Leverages LLMs for Generative Discovery of Thermally Activated Delayed Fluorescence (TADF) Molecular Candidates

July 12, 2026

ChemRxiv International

Overview

Firefly-Geni has been announced as an automated, interpretable active evolution framework integrating LLM-assisted data extraction, multitask property prediction, conditional molecule generation, and theoretical evaluation. This closed-loop workflow autonomously generates structurally novel Thermally Activated Delayed Fluorescence (TADF) molecular candidates by leveraging dispersed knowledge from scientific literature. It aims to accelerate the discovery of new materials for enhancing the efficiency of OLED displays and lighting.

In Depth

Key Findings

The ‘Firefly-Geni’ framework, an automated and interpretable active evolution system, has been unveiled, enabling the autonomous generation of Thermally Activated Delayed Fluorescence (TADF) molecular candidates through the strategic utilization of Large Language Models (LLMs). Firefly-Geni features a seamless closed-loop workflow that integrates LLM-driven data extraction, multitask property prediction, conditional molecule generation, and theoretical evaluation modules. This groundbreaking approach allows for the systematic leveraging of dispersed knowledge from scientific literature to efficiently explore novel TADF molecules with previously untapped structures.

Technical / Clinical Details

Firefly-Geni’s workflow comprises the following key steps: First, an LLM automatically extracts data related to TADF molecules, including structural information, synthesis protocols, and experimental results, from a vast corpus of scientific papers and databases. Next, a multitask property prediction model, trained on this extracted data, rapidly forecasts critical TADF properties (e.g., emission efficiency, stability, emission wavelength) of generated candidate molecules. Based on these predictions, a conditional molecule generation module designs and creates new molecular structures with desired property profiles. The generated molecules then undergo theoretical evaluation via quantum chemical calculations to rigorously verify their TADF performance. The results of this evaluation are fed back as learning data for the LLM in the subsequent exploration cycle, continuously improving the overall efficiency and accuracy of the framework. This iterative learning and generation loop allows Firefly-Geni to autonomously explore structurally diverse and high-performance TADF molecules that would be challenging to discover through conventional trial-and-error methods.

Background & Context

Thermally Activated Delayed Fluorescence (TADF) materials are garnering significant attention as next-generation emissive materials for organic light-emitting diode (OLED) displays and lighting, poised to replace traditional fluorescent and phosphorescent materials. TADF materials are essential for achieving energy-efficient OLED devices due to their ability to attain high internal quantum efficiency (IQE). However, designing stable and highly efficient TADF molecules has been exceptionally challenging, requiring a deep understanding of their complex molecular structures and excited-state physicochemical properties. Conventional TADF molecular exploration has largely depended on the experience and intuition of chemists, presenting a time-consuming and costly bottleneck. Autonomous discovery frameworks integrating LLMs, such as Firefly-Geni, hold the potential to break through this bottleneck and dramatically improve the speed and efficiency of TADF material development.

Strategic Significance & Outlook

The advent of Firefly-Geni is set to transform the R&D paradigm in the TADF materials field. Moving forward, this framework is expected to be applied not only to TADF molecules but also to the discovery of other functional molecules, such as solar cell materials, drug candidates, and catalysts. The research team aims to further enhance Firefly-Geni’s generative capabilities and interpretability, evolving its molecular design to account for more complex design requirements and synthetic feasibility. The widespread adoption of this technology is expected to significantly shorten the timeframe from new material discovery to practical application, leading to the proliferation of high-performance OLED devices and the advancement of sustainable energy technologies, ultimately bringing substantial benefits to society as a whole.

Source: https://chemrxiv.org/doi/10.26434/chemrxiv.15005548

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.