🌿 Agro AI Governance — IA para Gobernanza Agroalimentaria

Valentina Burbano Salazar, Josser Cordoba Rivas, Diana Carolina Argüello Casallas

Agro AI Governance es una plataforma open source de gobernanza agroalimentaria participativa con inteligencia artificial explicable. Convierte reportes del territorio, señales ciudadanas y evidencia comunitaria en decisiones priorizadas, trazables y auditables. La solución integra un bot de Telegram para captura de datos en tiempo real, un dashboard web para análisis y visualización, y una cadena de hashes SHA-256 inmutable para auditoría institucional.

El sistema fue desarrollado durante el Global South AI Safety Hackathon 2026 (19-21 de junio) con enfoque en equidad, transparencia y seguridad en el uso de IA para el sector agroalimentario de América Latina.

4. Contexto y Problemática

En América Latina, la toma de decisiones en el sector agroalimentario enfrenta múltiples desafíos: datos fragmentados, falta de trazabilidad, ausencia de mecanismos de participación ciudadana efectivos y brechas de equidad entre zonas rurales y urbanas. Las herramientas existentes se limitan a la visualización de datos sin ofrecer capacidades de priorización, auditoría o gobernanza verificable.

Agro AI Governance aborda esta problemática mediante un enfoque integral que combina:

Captura ciudadana descentralizada vía Telegram y web

Motor de reglas de gobernanza con 10 criterios de evaluación

Score de riesgo explicable (0-100) por registro

Cadena de auditoría inmutable para trazabilidad institucional

Auditoría de Impacto Dispar alineada con CONPES 4144 (Colombia, 2025)

5. Solución Propuesta

La plataforma opera en cuatro capas funcionales:

Capa 1 - Inteligencia Territorial

Motor de reglas con 10 criterios clasificados por severidad (Critical, High, Medium) que evalúa cada registro del CSV. Las reglas incluyen GOV-001 (trazabilidad de origen), GOV-002 (certificación para uso sensible), BIO-002 (umbral de contaminantes), entre otras. Cada registro recibe un score de riesgo de 0 a 100 y una clasificación en tres niveles: CONFIABLE, REVISIÓN o NO APTO.

Capa 2 - Participación Ciudadana

Bot de Telegram con 6 comandos interactivos (/start, /ayuda, /ultimo, /lotes, /registro, /catalogo) que permite a productores y gestores territoriales reportar descartes agroindustriales, adjuntar evidencia y recibir retroalimentación inmediata. El bot procesa archivos CSV de hasta 5,000 registros con barra de progreso en tiempo real.

Capa 3 - Gobernanza Verificable

Cadena de hashes SHA-256 encadenada donde cada evento (carga de lote, procesamiento, auditoría) registra un event_hash y un previous_hash. Una alteración no autorizada en la base de datos invalida la cadena completa, garantizando inmutabilidad y trazabilidad institucional.

Capa 4 - Equidad LATAM

Auditoría de Impacto Dispar que verifica que el ratio entre subgrupos regionales y poblaciones rurales/urbanas sea menor a 1.25, alineado con CONPES 4144 (Colombia, 2025) y el ASEAN Guide on AI Governance and Ethics. La calibración empírica valida que una brecha de cobertura mayor a 10 puntos porcentuales active revisión automática.

6. Arquitectura Técnica

Componente Tecnología

Backend Django 4.x

Base de datos SQLite

Bot de Telegram python-telegram-bot (long polling)

Motor de reglas Python puro - ruleset-2026-06-django-unificado-v1

Auditoría SHA-256 encadenado

Gráficas Matplotlib / charts.py

Analítica Jupyter Notebook + food_waste_charts.py

Despliegue Docker + docker-compose

7. Resultados y Métricas Alcanzadas

Durante el hackathon, el sistema fue probado con el dataset global_food_wastage_dataset.csv de 5,000 registros, obteniendo los siguientes resultados:

5,000 registros procesados en tiempo real

4,537 registros clasificados como CONFIABLE

407 registros en REVISIÓN (requieren inspección humana)

56 registros NO APTOS (errores estructurales o alertas críticas)

23 eventos registrados en la cadena de auditoría

Cadena de hashes validada como íntegra e inalterada

Auditoría de Impacto Dispar LATAM completada con ratio < 1.25

94% de riesgo territorial identificado

12 alertas activas

8 zonas monitoreadas

3 canales de reporte integrados (web, Telegram, analítica)

8. Alineación con Políticas Públicas

El proyecto está alineado con:

CONPES 4144 (Colombia, 2025) - Política de Equidad y Desarrollo Territorial

ASEAN Guide on AI Governance and Ethics - Principios de IA ética y gobernanza

Objetivos de Desarrollo Sostenible (ODS) - Hambre Cero y Reducción de Desigualdades

9. Repositorio y Documentación

Repositorio GitHub:

https://github.com/BurbanoValentina/agro_governance_unified

Documentación completa disponible en Google Drive:

Video de presentación del proyecto (pitch)

Diapositivas de la solución

Documentación técnica y manual de usuario

Base de datos con registros de prueba

Podcast explicativo del proyecto

Notebook Jupyter de auditoría de equidad alimentaria

10. Equipo

Desarrollado por BurbanoValentina para el Global South AI Safety Hackathon 2026.

11. Uso de IA (LLM Statement)

Claude (Anthropic) fue utilizado para asistir en el encuadre del proyecto, redacción de secciones y sugerencias de estructura de código. Todos los resultados cuantitativos, decisiones metodológicas y afirmaciones de política fueron especificados y verificados de forma independiente por el equipo.

Reviewer's Comments

Reviewer's Comments

Arrow
Arrow
Arrow

No reviews are available yet

Cite this work

@misc {

title={

(HckPrj) 🌿 Agro AI Governance — IA para Gobernanza Agroalimentaria

},

author={

Valentina Burbano Salazar, Josser Cordoba Rivas, Diana Carolina Argüello Casallas

},

date={

},

organization={Apart Research},

note={Research submission to the research sprint hosted by Apart.},

howpublished={https://apartresearch.com}

}

Recent Projects

OliGraph: graph-based screening of large oligopools

Existing synthesis screening tools cannot evaluate short oligonucleotide pools, whose overlapping fragments can be reassembled into regulated sequences via polymerase cycling assembly (PCA) yet fall below gene-length detection thresholds. We present OliGraph, an open-source tool that constructs a bi-directed overlap graph from an oligonucleotide pool and extracts contigs for downstream gene-length screening. An optional PCA mode retains only cross-strand overlaps consistent with PCA chemistry. We validated OliGraph in a blinded study across ten simulated pools (70–9,184 oligonucleotides, 30–300 bp) spanning four risk categories. BLAST screening of individual oligonucleotides failed to identify sequences of concern in most pools: three returned zero hits, and vector noise obscured true positives in the remainder. After OliGraph assembly, contig-level BLAST matched the longest assembled sequences (up to 1,905 bp) to sequences of concern at 97–100% identity. In one pool, assembly collapsed 1,634 individual BLAST results into 10 hits from a single contig, all assigned to the same source organism. PCA mode correctly distinguished assemblable from non-assemblable fragments within the same pool. Two pools with no assemblable structure yielded no contigs. OliGraph processed all pools in under 0.2 seconds, fast enough for real-time order screening and consistent with proposals to bring oligonucleotide orders within the scope of synthesis screening regulation.

Read More

BioRT-Bench: A Multi-Attack Red-Teaming Benchmark for Bio-Misuse Safeguards in Frontier LLMs

Frontier AI laboratories are expected to maintain safeguards against biological misuse, but whether deployed models actually refuse bio-misuse queries under adversarial pressure is largely unmeasured in the public literature. We introduce BioRT-Bench, a benchmark that runs four attack methods (direct request, PAIR, Crescendo, and base64 encoding) against four frontier models (Claude Sonnet 4.6, GPT-5.4, DeepSeek V4-flash, Kimi K2.5) across 40 prompts spanning five biosecurity-relevant categories. Responses are scored by a calibrated judge extending StrongREJECT with two bio-specific dimensions: specificity and actionability. We measure Attack Success Rate (ASR), where 0 means the model fully refused and 1 means it provided specific, actionable bio-misuse content. Our results reveal a sharp robustness divide: Chinese frontier models (DeepSeek, Kimi) have under 5% refusal rates even under direct request (ASR 0.88 and 0.79), while Western models (Claude, GPT) maintain substantially stronger safeguards (ASR 0.15 and 0.16). Crescendo is the most effective attack across all models, both in bypassing refusal and in eliciting actionable content. Claude Sonnet 4.6 is the most robust model tested, achieving 100% refusal against base64-encoded prompts.

Read More

PROTEUS (PROTein Evaluation for Unusual Sequences): Structure-Informed Safety Screening for de novo and Evasion-Prone Protein-Coding Sequences

AI protein design tools like RFdiffusion, ProteinMPNN, and Bindcraft make it trivial to produce low-homology sequences that fold into active, potentially hazardous architectures. However, sequence homology-based biosafety screening tools cannot detect proteins that pose functional risk through structurally novel mechanisms with no sequence precedent. We present a tiered computational pipeline that addresses this gap by combining MMseqs2 sequence alignment with structure-based comparison via FoldSeek and DALI against curated toxin databases totaling ~34,000 entries. AlphaFold2-predicted structures are screened for both global fold similarity (FoldSeek) and local active/allosteric site geometry (DALI), capturing convergent functional hazards that sequence screening misses. The pipeline was validated against a panel of toxins, benign proteins, structural mimics, and de novo-designed Munc13 binders, as well as modified ricin variants with residue substitutions. We additionally tested robustness to partial-synthesis evasion, where a bad actor submits multiple shorter coding sequences intended for downstream reassembly into a full toxin-coding gene. We found that while sequence-based screening did not identify any de novo ricin analogues with high certainty, the combined pipeline with FoldSeek and DALI identified all 24 tested de novo ricins as toxic.

Read More

OliGraph: graph-based screening of large oligopools

Existing synthesis screening tools cannot evaluate short oligonucleotide pools, whose overlapping fragments can be reassembled into regulated sequences via polymerase cycling assembly (PCA) yet fall below gene-length detection thresholds. We present OliGraph, an open-source tool that constructs a bi-directed overlap graph from an oligonucleotide pool and extracts contigs for downstream gene-length screening. An optional PCA mode retains only cross-strand overlaps consistent with PCA chemistry. We validated OliGraph in a blinded study across ten simulated pools (70–9,184 oligonucleotides, 30–300 bp) spanning four risk categories. BLAST screening of individual oligonucleotides failed to identify sequences of concern in most pools: three returned zero hits, and vector noise obscured true positives in the remainder. After OliGraph assembly, contig-level BLAST matched the longest assembled sequences (up to 1,905 bp) to sequences of concern at 97–100% identity. In one pool, assembly collapsed 1,634 individual BLAST results into 10 hits from a single contig, all assigned to the same source organism. PCA mode correctly distinguished assemblable from non-assemblable fragments within the same pool. Two pools with no assemblable structure yielded no contigs. OliGraph processed all pools in under 0.2 seconds, fast enough for real-time order screening and consistent with proposals to bring oligonucleotide orders within the scope of synthesis screening regulation.

Read More

BioRT-Bench: A Multi-Attack Red-Teaming Benchmark for Bio-Misuse Safeguards in Frontier LLMs

Frontier AI laboratories are expected to maintain safeguards against biological misuse, but whether deployed models actually refuse bio-misuse queries under adversarial pressure is largely unmeasured in the public literature. We introduce BioRT-Bench, a benchmark that runs four attack methods (direct request, PAIR, Crescendo, and base64 encoding) against four frontier models (Claude Sonnet 4.6, GPT-5.4, DeepSeek V4-flash, Kimi K2.5) across 40 prompts spanning five biosecurity-relevant categories. Responses are scored by a calibrated judge extending StrongREJECT with two bio-specific dimensions: specificity and actionability. We measure Attack Success Rate (ASR), where 0 means the model fully refused and 1 means it provided specific, actionable bio-misuse content. Our results reveal a sharp robustness divide: Chinese frontier models (DeepSeek, Kimi) have under 5% refusal rates even under direct request (ASR 0.88 and 0.79), while Western models (Claude, GPT) maintain substantially stronger safeguards (ASR 0.15 and 0.16). Crescendo is the most effective attack across all models, both in bypassing refusal and in eliciting actionable content. Claude Sonnet 4.6 is the most robust model tested, achieving 100% refusal against base64-encoded prompts.

Read More

This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.