MSc / PhD applicant 2027 cohort · open to fully funded positions
Computational Agricultural Scientist · Plant Pathology & Bioinformatics · One Health trajectory
Final-year B.Sc. at EARTH University in Costa Rica seeking graduate research positions in computational plant science, food systems innovation, and One Health. My undergraduate work bridges deep learning, bioinformatics, and tropical agricultural science.
Français ·
Kreyòl ·
English ·
Español
Python
R
PyTorch
TensorFlow
Pandas
NumPy
scikit-learn
QGIS
LaTeX
I am a final-year agricultural sciences student at EARTH University in Costa Rica, applying for graduate research in computational plant science, food systems innovation, and One Health. My undergraduate work has built a deliberate bridge between field-grounded agricultural science and data-driven modeling, and I want to continue that bridge at the MSc and PhD level.
I came to this research from two directions. From Haiti, where I grew up watching coffee, mango, and breadfruit production constrained by disease, post-harvest loss, and a fragile food system. From Costa Rica, where EARTH's focus on tropical agriculture met my own curiosity about machine learning. The intersection is the territory I want to live in as a scientist: how can computational tools make tropical food systems more resilient, more productive, and more legible to the people who depend on them?
Three threads in particular pull at me:
These three threads rest on a shared methodological commitment: reproducible scientific computing in low-resource settings. RflowLabs and Mi Salvador are early outputs of this commitment, open tools, transparent pipelines, every result reachable from a single seed.
The graduate position I am seeking is one where I can sharpen these instincts under a supervisor working at the intersection of AI, biology, and agriculture, and contribute to research that is at once methodologically rigorous and grounded in the realities of tropical food systems.
The research I am bringing into graduate study. Each is grounded in a real problem and a real population, and each is the seed of a longer programme.
A full-stack browser-based laboratory for teaching and computing genetics and bioinformatics. Five major modules, an LLM-powered tutor, four languages, and reproducible-from-a-seed simulation throughout.
Designed for genetics and bioinformatics courses where students need to touch real molecular tools without leaving the classroom or installing a desktop bioinformatics stack. The AI tutor extends office hours into the browser. Multilingual (English, French, Spanish, Haitian Creole), themed dark / light, fully reproducible: every simulation run is seedable and every result downloadable as a trace. To date, Mi Salvador has been used by more than 62 users across the Americas, Europe, and Africa.
A deep-learning study applying knowledge-distilled vision transformers (ViT) to ordinal severity grading of Hemileia vastatrix (Roya) and Mycena citricolor (Ojo de Gallo) on field-captured coffee leaves.
Knowledge distillation transfers learned representations from a large pretrained ViT teacher into a smaller student network suitable for smartphone-class deployment. Ordinal regression heads preserve the natural ranking of severity classes, a structure that standard cross-entropy discards. Coffee leaf rust alone has cost Latin American producers an estimated USD 3.2 billion since the 2012 outbreak (Avelino et al., 2015); this work targets smallholder-deployable diagnostics. Manuscript in preparation, first author. Target venues: Computers and Electronics in Agriculture or Agriculture.
1,400 leaves · iPhone 11 + 3D-printed rig · ColorChecker · 4,200 images at 0°/45°/abaxial.
EMSAM → CVAT → STAPLE consensus · 3 phytopathologists · 10% triplicate subset · Krippendorff α ≥ 0.80.
Method A: SegFormer-B3 (45M) · Method B: FastViT-T8 (4M) with DINOv2-LoRA teacher and CORN ordinal head.
3-fold grouped CV · BCa paired bootstrap · TOST equivalence · ±0.05 margin · power 0.97 at Δκ = 0.
FastViT to LiteRT (Android) / Core ML (iOS) · on-device inference, fully offline · 113× fewer FLOPs.
5 architectures plotted on the params–κ plane. The 4 M-parameter FastViT-T8 is hypothesised to match the 45 M SegFormer-B3 to within ±0.05 κ.
Five preregistered floors; all reported regardless of outcome. Bars show target value on the 0–1 scale.
Two One-Sided Tests (TOST) with equivalence margin ±0.05. Expected: 95% CI of Δκ falls fully within the shaded zone.
Probability of correctly declaring equivalence at N = 1,400. Power = 0.97 at Δ = 0, dropping outside ±0.05.
A first working system for the thesis' second target disease, Mycena citricolor (ojo de gallo). It segments necrotic lesions pixel by pixel, measures severity as lesion area ÷ leaf area, and assigns a G0–G3 grade, running fully offline on a phone.
Seven segmentation models plus a transformer classifier, all on the same held-out test set, ranked by lesion Dice. The deployed 17 MB model sits third, behind only a 106 MB and a 324 MB network.
| Model | Grade | QWK | Dice | IoU | Size | On-device |
|---|---|---|---|---|---|---|
| SegFormer-B5 (1024px) | 100% | 1.00 | 0.940 | 0.916 | 324 MB | · |
| DeepLabV3+ ResNet-50 (1024px) | 100% | 1.00 | 0.933 | 0.906 | 106 MB | · |
| DeepLabV3+ MobileNetV2 (1024px) | 100% | 1.00 | 0.924 | 0.895 | 17 MB | Yes |
| SegFormer-B5 (512px) | 100% | 1.00 | 0.911 | 0.878 | 324 MB | · |
| DeepLabV3+ EfficientNet-B7 (512px) | 96.4% | 0.944 | 0.893 | 0.851 | 242 MB | · |
| DeepLabV3+ MobileNetV2 (512px) | 98.2% | 0.972 | 0.892 | 0.850 | 17 MB | Yes |
| DeepLabV3+ ResNet-50 (512px) | 98.2% | 0.972 | 0.891 | 0.856 | ~100 MB | · |
| PLA-ViT (ViT-B/16 classifier) | 98.2% | 0.972 | n/a | n/a | 328 MB | · |
Per-leaf metrics on the real held-out test set (worst-face aggregation). Grade accuracy saturates at 100% for several models, so lesion Dice and model size are the true differentiators. PLA-ViT is a whole-leaf classifier and produces no lesion map.
A decision-support research tool for coffee-leaf disease assessment, not a validated clinical diagnosis. Reported metrics are on flatbed scans; phone photos yield estimates.
Pick one of my sample leaves, or upload your own coffee-leaf photo, and the segmentation model runs entirely on your device via ONNX Runtime Web. It localizes the ojo de gallo (Mycena citricolor) lesions, measures the affected leaf area, and returns an ordinal severity grade (G0–G3). No image ever leaves your browser.
Pick a sample leaf to begin, the model downloads once, then runs offline.
A season's water-quality, yield, and fish-health data from a coupled aquaponic system, collected as part of the integrated aquaponics research program at Auburn's E.W. Shell Fisheries Center.
Designed and operated biometric monitoring protocols and contributed to data pipelines for real-time system performance evaluation. The dataset, pH, ammonia/nitrite/nitrate cycling, dissolved oxygen, fish length-weight, tomato yield, is the seed for the digital-twin modeling I want to pursue at the graduate level.
A closed-loop pH stabilizer for nutrient-film hydroponics, dosing dilute acid and base via peristaltic pumps in response to a live probe reading on an Arduino board.
Hydroponic crops sit in a narrow pH window (5.5 – 6.5); drift outside that window costs yield within hours. The prototype reads a glass pH probe through an analog converter, runs a debounced setpoint loop on the Arduino, and triggers two peristaltic pumps (acid / base) in short pulses with a cool-down period to prevent over-shooting. An LCD shows the live reading and the last action. Built and tested against laboratory pH standards over a weekend, then submitted to the Cenfotec inter-university Arduino competition where it placed 6th of 16 projects.
A field-and-classroom role at the Geomatics & Remote Sensing Center: leading topographic survey campaigns for clients on and off campus, and teaching the GIS workflows that come after the data is collected.
As team leader I plan and run levantamientos (topographic surveys) for real clients (university plots, neighbouring farms, school grounds) using total stations and GNSS receivers, then take the data into QGIS for cleaning, georeferencing, and map production. On the teaching side I have led GIS sessions for more than 85 students across first-year and second-year EARTH cohorts plus visiting high-school groups, covering everything from coordinate systems and projection choice to field-data import and cartographic output. To date the team has mapped over 6 hectares of campus and adjacent terrain.
The full application packet. Please feel free to circulate any of these to colleagues or fellow committee members.
Each of the following has agreed to write on my behalf. Please contact me before reaching out so they can prepare a letter tuned to your program; for casual queries, direct contact details are provided.
Where you would find me on a Tuesday morning. EARTH University's main campus sits in the Caribbean lowlands of Limón, surrounded by the tropical forestry plots and aquaponics greenhouses my work draws on.