Literature-grounded LLM scientist
Retrieves evidence, extracts leaf and formulation descriptors, critiques hypotheses, checks constraints, and prepares auditable rationales for human review.
Programming leaf-surface delivery with an open AI discovery platform.
FoliarShield-AI turns foliar delivery from trial-and-error formulation into an evidence-linked platform for retention, rainfastness, controlled release, and living-payload survival on difficult-to-wet leaves.
Evidence graph
Bayesian loop
LLM scientist
Wet-lab validation
The challenge
Foliar delivery often fails after the droplet reaches the leaf.
Sprayed droplets and encapsulated payloads rebound, splash, roll off, evaporate, wash away, or lose activity under UV, drying, and rainfall. The problem is sharpest on waxy, hydrophobic, or otherwise difficult-to-wet crop leaves.
Current formulation practice rarely couples droplet impact, spray mode, leaf wettability, rheology, encapsulation, release kinetics, and microbial survival inside one predictive design framework.
Experimental base
Built around measurable droplet, encapsulation, and leaf-surface assays.
Platform architecture
A scientist-in-the-loop system for leaf-surface delivery design.
Evidence-linked knowledge graph
Connects leaf traits, droplet physics, capsule architecture, rheology, release kinetics, microbial persistence, and evidence provenance.
Batched Bayesian active learning
Uses uncertainty and expected Pareto improvement to recommend formulation batches under safety, sprayability, feasibility, and material constraints.
Tiered experimental assays
Fast image-based droplet, spray, rainfastness, evaporation, and early-release assays drive iteration; slower viability and plant assays validate shortlists.
Expected outcomes
From leaf assays to reusable public infrastructure.
2-3x
Target hit-rate lift for candidates meeting retention, rainfastness, release, and viability windows.
25-30%
Target reduction in screening cost per viable candidate compared with baseline exploration.
10+
External research groups served through datasets, tools, benchmark tasks, workshops, and tutorials.
Open
Knowledge graph, assay records, image workflows, benchmark suites, model cards, and negative results.
Discovery loop
Fast physics labels guide the loop; biology validates the shortlists.
Viscosity, oil fraction, shell chemistry, capsule size, surfactant/adjuvant choices, release half-life, and payload class.
Droplet impact videos, contact-line motion, spray coverage, retained fluorescent intensity, wash-off, evaporation, and early release.
Release curves, microbial viability after atomization and drying, UV exposure, recovery, and controlled polyhouse or plant validation.
Beneficiaries
Designed for climate-vulnerable agriculture and public-interest bioinput research.
Smallholder farmers facing heat, drought, erratic rainfall, and rising input costs
Crop-protection, formulation, wetting, and agricultural microbiology researchers
Public institutes and ICAR-linked partners evaluating foliar delivery systems
Mission-driven bioinput developers building lower-input crop-management tools
Roadmap
A staged path from benchmark assays to open scientific infrastructure.
Benchmark foundation
Finalize the rice and waxy-brassica leaf panel, payloads, responsible-use protocols, knowledge graph, baseline assays, and metrics.
Nonliving payload loop
Run fluorescent tracer or low-risk surrogate batches through droplet, spray, wash-off, evaporation, release, and Bayesian optimization cycles.
Coupled retention-release design
Expand to spray-mode transfer, rheology, encapsulation variables, uncertainty calibration, literature-grounding evaluation, and compatibility screening.
Living-payload validation
Introduce a non-pathogenic Bacillus-like payload and compare unencapsulated, conventional carrier, and AI-selected cloaked or capsule systems.
Open release and adoption
Complete evaluation, package datasets and code, publish methods and failures, run workshops, and formalize community governance.
