Spectrometric Identification of Phytochemicals in Ocimum sanctum (Tulsi) Leaves Using UV-Visible Spectroscopy and Generative AI (Gemini) Interpretation

Abstract

Background: The standardization of herbal medicines, particularly Ocimum sanctum (Tulsi), is frequently hindered by the "quality gap" in resource-limited settings. While gas chromatography-mass spectrometry (GC-MS) represents the gold standard for phytochemical profiling, its prohibitive cost renders it inaccessible for routine analysis in developing regions. Conversely, UV-Visible spectrophotometry is cost-effective but historically lacks the specificity required for qualitative structural elucidation.

Objective: This study investigates a novel "Frugal Science" methodology, termed "Prompt-Based Spectrometry," which integrates the hardware accessibility of UV-Visible spectrophotometry with the semantic reasoning capabilities of Generative Artificial Intelligence (Google Gemini) to democratize phytochemical identification.

Methods: Methanolic leaf extracts of Ocimum sanctum were prepared via cold maceration and scanned between 200 nm and 800 nm. The resulting spectral data was interpreted using a structured "Chain-of-Thought" prompting protocol, where Google Gemini functioned as a "Virtual Spectroscopist" to correlate absorption maxima (λmax) with specific electronic transitions. The AI-generated predictions were subsequently validated using standard qualitative chemical screening (Ferric Chloride, Shinoda, and Salkowski tests).

Results: The spectral analysis exhibited distinct absorption maxima at 215 nm, 268 nm, 282 nm, and 327 nm. The AI model successfully deconvoluted these signals, tentatively identifying Ursolic Acid (triterpenoid), Apigenin (flavone), Eugenol (phenylpropanoid), and Rosmarinic Acid (phenolic acid). These predictions showed a 90% concordance with wet-lab validation assays, which confirmed the presence of phenolic and flavonoid scaffolds.

Conclusion: This research demonstrates that Large Language Models (LLMs) can effectively serve as zero-cost "virtual reference libraries," capable of decoding complex spectral fingerprints. The proposed method offers a rapid, accessible alternative for the preliminary quality control of herbal raw materials.