Artificial Intelligence and Computer Vision Engineer focused on designing, training, and deploying production-grade visual perception systems, statistical machine learning pipelines, and interactive spatial interfaces.
My engineering work spans the complete lifecycle of intelligent systems: from dataset curation, leak-free preprocessing, and multi-stage convolutional fine-tuning (PyTorch, TensorFlow / Keras 3, Scikit-Learn) to low-latency edge quantization (Raspberry Pi 5 + Hailo-8L 13 TOPS NPU), asynchronous inference APIs (FastAPI), and 3D digital twin visualization (Three.js / WebGL).
NAME Alfredo Ramos
DISCIPLINE Artificial Intelligence, Computer Vision & Systems Engineering
PRIMARY STACK Python, PyTorch, TensorFlow, Keras 3, Scikit-Learn, OpenCV, FastAPI, Three.js, TypeScript
EDGE TARGETS Raspberry Pi 5, Hailo-8L NPU (13 TOPS INT8), Apple Silicon MPS, Docker Containers
METHODOLOGY Reproducible pipelines, cost-sensitive recall calibration, explainable AI (Grad-CAM)
| System | Domain | Core Architecture & Stack | Empirical Benchmarks & Capabilities |
|---|---|---|---|
| inspeccion-industrial | Deep Learning & Quality Control |
TensorFlow 2.20 Keras 3 MobileNetV2 Streamlit Playwright
|
Automated binary defect detection across 7,348 submersible pump impeller castings. Evaluates 5 progressive architectures from dense MLP to MobileNetV2 fine-tuning, achieving 99.72% test accuracy and 99.62% defect recall (713/715 test castings). |
| peoyecto-retail | Edge Computer Vision & 3D Twin |
PyTorch YOLO11x RT-DETR RF-DETR Hailo-8L FastAPI Three.js
|
Autonomous retail shelf inventory system trained on SKU-110K (1.7M bounding boxes). Combines multi-detector Weighted Boxes Fusion (98.5% mAP) with 30 FPS INT8 edge execution on Raspberry Pi 5 and a standalone Three.js 3D store simulator. |
| defect-vision-studio | Explainable AI (XAI) |
Python Grad-CAM FastAPI OpenCV HTML5 Canvas
|
Spatial interpretability engine implementing Gradient-weighted Class Activation Mapping ($L_{\text{Grad-CAM}}^c = \text{ReLU}(\sum_k \alpha_k^c A^k)$) for real-time defect hotspot localization (12.0 ms mean latency) across REST, CLI, and web interfaces. |
| credit-risk-ml-pipeline | Statistical Machine Learning |
Scikit-Learn ColumnTransformer GridSearchCV PCA K-Means
|
Leak-free credit default probability pipeline on the German Credit dataset (N = 1,000). Compares Logistic Regression, RBF SVM, Random Forest, and Decision Trees via 5-fold Stratified CV (0.8092 test ROC-AUC), paired with 2D PCA and K-Means segmentation. |
| neural-style-transfer-api | Deep Learning Synthesis |
PyTorch VGG19 FastAPI JavaScript
|
Production neural style transfer service leveraging intermediate VGG19 perceptual feature representations and Gram matrix optimization, exposed via an asynchronous FastAPI server and interactive web workbench. |
| conaempleo-showcase | Platform Architecture |
JavaScript CSS3 System Design Security
|
Sanitized public reference architecture of the CONAEMPLEO institutional employability platform, demonstrating multi-role workflows, bilingual interface systems, and strict data-isolation boundaries. |
| Layer | Technologies & Methodologies |
|---|---|
| Deep Learning & Computer Vision | PyTorch · TensorFlow 2.20 · Keras 3 · OpenCV · YOLO11 · RT-DETR · RF-DETR · MobileNetV2 · ResNet-18/50 · VGG19 · RetinaNet FPN · Weighted Boxes Fusion (WBF) · Grad-CAM Saliency · Perspective Homography |
| Machine Learning & Data Science | Scikit-Learn · NumPy · Pandas · Matplotlib · Seaborn · Logistic Regression · Support Vector Machines (RBF) · Random Forest · K-Means Clustering · Principal Component Analysis (PCA) · Stratified K-Fold CV · GridSearchCV |
| Edge AI, Backend & Infrastructure | Raspberry Pi 5 · Hailo-8L NPU (13 TOPS) · INT8 Post-Training Quantization · FastAPI · Uvicorn · Node.js · Streamlit · Gradio · Docker · Docker Compose · Linux / Bash · Git |
| Spatial Web, Frontend & Tooling | JavaScript (ES2024) · TypeScript · Three.js · WebGL · HTML5 Canvas · React · Tailwind CSS · CSS3 Architecture · GSAP · Playwright Automated Testing · CLI Systems |
- inspeccion-industrial — Automated industrial casting defect inspection via MobileNetV2 fine-tuning (
99.72%accuracy,99.62%recall). - peoyecto-retail — Edge-accelerated retail inventory computer vision system (
YOLO11+RT-DETR+Hailo-8L NPU) and Three.js 3D store twin. - defect-vision-studio — Explainable AI (
Grad-CAM) defect saliency engine with FastAPI backend, CLI utility, and interactive inspection workbench. - credit-risk-ml-pipeline — End-to-end credit risk scoring and borrower segmentation pipeline using Scikit-Learn, Stratified
GridSearchCV,PCA, andK-Means. - neural-style-transfer-api — Deep convolutional style transfer engine powered by
PyTorch,VGG19, andFastAPI. - vision-document-scanner — Document contour detection, Canny edge extraction, and 4-point perspective homography pipeline.
- signalforge-ai — Time-series telemetry anomaly detection and incident diagnostics laboratory.
- clusternebula — Interactive 3D data science laboratory visualizing
K-Meansclustering trajectories inThree.js/WebGL.
- ia-oculta — Security command-line scanner detecting hidden prompt injections, zero-width Unicode steganography, and adversarial LLM payloads.
- repodoctor — Automated repository auditing CLI verifying documentation completeness, structural hygiene, and secret exposure risks.
- promptframe — Deterministic prompt engineering workbench for structuring system instructions, constraints, and evaluation rubrics.
- flowfixer — Structured debugging diagnostic engine mapping runtime exceptions to deterministic resolution workflows.
- briefsmith — Technical specification generator transforming product requirements into scoped architecture roadmaps.
- lifepilot — Dual-surface (Web + Terminal CLI) scheduling and task-prioritization engine.
- conaempleo-showcase — Privacy-safe architectural case study and bilingual frontend implementation for the CONAEMPLEO employability platform.
- motionforge — Visual
GSAPkinetic choreography builder for scroll-triggered sequences and production code generation. - aura-motion-kit — Micro-interaction design system and responsive CSS motion primitives.
- excel-maniobras-descarga — Standardized operational workflow and spreadsheet automation templates for logistics maneuver tracking.
- LinkedIn: linkedin.com/in/alfredo-ramos-olivan-07751a3a5
- Email: alfredooliva765498@gmail.com
- GitHub: github.com/Alfredo-ctrl