AMD DEVELOPER HACKATHON · UNICORN TRACK

Designing the future of
biopesticides

An AI pipeline that converts a farmer's pest report into ranked, safety-checked dsRNA candidates — in minutes, not months. Powered by local Llama 3.2 3B, PyTorch, and a 14-species safety panel.

TeamBiopesticide-AI
DateJune 2026
LicenseMIT
Compute100% Local
Biopesticide-AI

The problem

Chemical pesticides are failing on three fronts simultaneously

$84B
Annual market

Resistance collapse

600+ arthropod species have documented resistance. Brown planthopper alone destroys 30% of Asian rice yields in outbreak years. Farmers apply higher doses that accelerate resistance selection.

Decades
Environmental persistence

Ecological damage

Neonicotinoids linked to pollinator collapse. EU banned outdoor use in 2018. Soil and water contamination persists for years after application. Regulators are tightening globally.

3-6 mo
Design loop

Expert-gated RNAi

RNAi biopesticides are the new chemistry (EPA registered Ledprona in 2023), but dsRNA design takes 3-6 months per target gene. No integrated safety, fate, or regulatory layer exists.

02
Biopesticide-AI

The solution

Compress the dsRNA design loop from months to minutes

An end-to-end pipeline that any farmer or agronomist can run from a laptop.

Select a pest target. The pipeline tiles pest transcripts into 200-nt dsRNA precursors, dices them into 21-nt siRNAs, scores each for efficacy, checks off-target risk against 14 non-target species, predicts environmental half-life, and generates safety cards + regulatory memos.

7
Pest targets
14
Safety species
~4 min
Design loop
$0
Per design
03
Biopesticide-AI

How it works

6-stage pipeline from pest selection to regulatory memo

01

Select

Click a pest target card (7 species supported) or describe the problem in natural language.

02

Tile & Dice

PyTorch backend tiles pest transcripts into 200-nt dsRNA precursors, then dices into 21-nt siRNAs.

03

Score efficacy

Dilated CNN (HyenaDNA-inspired) scores each siRNA. Caduceus-Ph-1 adapter available for SOTA accuracy.

04

Check safety

K-mer index checks off-target risk against 14 non-target species (pollinators, livestock, aquatic, human).

05

Predict fate

Physics-Informed Neural Network predicts environmental half-life based on sequence and field conditions.

06

Simulate & report

1000-trial Monte Carlo wet-lab simulation. Llama 3.2 3B generates safety cards + EPA-style regulatory memo.

04
Biopesticide-AI

Architecture

PyTorch + ROCm-ready + local Ollama LLM

Data flow

Pest card selection
Tile 200-nt precursors → Dice 21-nt siRNAs
Dilated CNN
efficacy score
K-mer index
14-species safety
PINN
half-life
Learned ranker → final score
Monte Carlo wet-lab simulation (1000 trials)
Ollama Llama 3.2 3B → safety cards + regulatory memo

Tech stack

PyTorch 2.4ROCm-ready
Caduceus-Ph-1SOTA DNA model
Dilated CNNHyenaDNA-inspired
Ollama 3.2 3BLocal LLM
FastAPIREST backend
Chart.jsAnalytics frontend
DockerContainerized
MITOpen source
05
Biopesticide-AI

Species coverage

7 pest targets · 14-species safety panel covering every ecological role regulators evaluate

7 Pest targets

Brown planthopper
N. lugens · Rice
Fall armyworm
S. frugiperda · Maize
Desert locust
S. gregaria · Wheat
Stem borer
C. suppressalis · Rice
Peach-potato aphid
M. persicae · Veg
Potato beetle
L. decemlineata · Potato
Tobacco whitefly
B. tabaci · Tomato

14 Safety panel species

Honeybee
A. mellifera · Pollinator
Bumblebee
B. terrestris · Pollinator
Leafcutter bee
M. rotundata · Pollinator
Ladybug
A. bipunctata · Predator
Lacewing
C. carnea · Predator
Earthworm
E. fetida · Soil
Water flea
D. magna · Aquatic
Cattle
B. taurus · Livestock
Zebu
B. indicus · Livestock
Chicken
G. gallus · Poultry
Sheep
O. aries · Livestock
Pig
S. scrofa · Livestock
Zebrafish
D. rerio · Aquatic
Human
H. sapiens · Safety
06
Biopesticide-AI

Live demo results

Brown planthopper · 180 siRNAs scored · results in 0.08 seconds

Top 5 candidates

ACAATGAGGTACAGATGTATA Eff: 84% · OT: 0.000 · HL: 25.5h
TGACGTCCGTAGGCCTTAACC Eff: 82% · OT: 0.000 · HL: 37.7h
ACATAAGAATTAATATCTAAA Eff: 78% · OT: 0.000 · HL: 24.0h
AGGGCGGACTTCAGGTGTTGT Eff: 74% · OT: 0.000 · HL: 37.7h
ACAGGCGGCGGTAGCTTGTAA Eff: 69% · OT: 0.000 · HL: 37.7h

Pipeline stats

Design loop time0.08 seconds
Pest transcripts loaded5
dsRNA precursors tiled20
siRNAs scored180
Safety panel species14
Off-target max (all candidates)0.000
Cost per design$0.0000
Compute location100% local
07
Biopesticide-AI

Virtual wet-lab simulation

1000-trial Monte Carlo · 6-stage cellular knockdown pipeline · literature-informed kinetics

Stage 1

Delivery

Lipofection efficiency for in vitro screens. Longer PINN half-life improves delivery stability.

85-95%
Stage 2

Uptake

Cellular uptake efficiency. GC-content dependent (optimal at 45% GC).

75-95%
Stage 3

Dicer processing

Dicer processes dsRNA into 21-nt siRNAs. Reynolds score dependent, repeats penalized.

80-95%
Stage 4

RISC loading

Thermodynamic asymmetry determines guide vs passenger strand. Reynolds rules 3-7.

70-95%
Stage 5

Target cleavage

mRNA cleavage rate. CNN-predicted efficacy is the primary driver.

20-98%
Stage 6

Phenotype

Soft-saturation dose-response. Reports mean KD, 95% CI, P(KD>70%).

38-51%
08
Biopesticide-AI

Production upgrade path

Real measured data + AMD GPU training · architecture is production-ready, only training data is synthetic

ComponentDemo (synthetic)Production (real data + AMD GPU)
Training set1,000 synthetic siRNAs30,000+ measured siRNAs (siRecords)
Safety panel14 synthetic transcriptomes14 real NCBI transcriptomes (~2 GB)
ModelDilated CNN, 2 min on CPUCaduceus fine-tune, 15 min on MI250
Validation AUC1.00 (trivially separable)0.82-0.92 (realistic)
Off-target accuracy0% (random sequences)Real per-species risk scores
Training cost$0 (local CPU)~$1.50 (1 hour MI250)

Codebase already supports production via three CLI flags: --source real · --device cuda · --use-caduceus

09
Biopesticide-AI

Business case

$84B market · EPA regulatory precedent set · no competitors with integrated AI pipeline

Market opportunity

$84B
Total pesticide market (TAM)
$12B
Biopesticide segment (SAM, 15% CAGR)
$480M
RNAi biopesticides by 2030 (SOM)
2023
EPA registered Ledprona (first dsRNA)

Competitive landscape

Greenlight Biosciences2-4 weeks
AgroSpheres2-3 weeks
RNAissance Ag3-6 weeks
Academic tools (DeepRiPE)1-2 weeks
Biopesticide-AI< 4 minutes

Only platform combining SOTA sequence models, PINN fate prediction, 14-species safety panel, and LLM-generated regulatory docs. 100-1000x faster than competitors.

10
Biopesticide-AI

Roadmap

From hackathon MVP to Series A in 12 months

0-3 months

BPH MVP

  • Train on real siRecords data (30k siRNAs)
  • Customer discovery with 3 Indian rice cooperatives
  • Validate against published BPH RNAi literature
  • Deploy on AMD MI250 via ROCm 6.1
3-6 months

Multi-pest expansion

  • Extend to fall armyworm (maize)
  • Extend to desert locust (wheat)
  • Add GNN for seed-region off-target modeling
  • Partner with university ag biotech labs
6-12 months

Series A

  • Raise $5M Series A
  • Wet-lab partnership (TNAU or UC Davis)
  • Expand safety panel to 30+ species
  • EPA FIFRA Section 3 registration filing
11

Building the first
LLM-native biopesticide company

The hackathon is the proving ground. The market is $84 billion. The regulatory pathway is open. The technology works. We are ready to build what's next on AMD.

$84B
Market
4 min
Design loop
14 sp.
Safety panel
$0
Per design
MIT licensed · AMD Developer Hackathon Unicorn Track · June 2026