Machine Learning Engineer | PhD in Artificial Intelligence | Open Source Enthusiast
I bridge the gap between academic research and production-grade ML systems. Currently working as an AI Engineer @ Unicaja, I specialize in Computer Vision, Deep Learning, and Robotics. My background spans from low-level electronics to high-level AI architectures.
- 🎓 PhD in Artificial Intelligence, MSc in Informatics, BSc in Electronics Engineering.
- 🔭 Working on: Realtime voice agents, conversational AI, computer vision.
- 🏠 Based in: Gijón, Spain (🇪🇸).
- 💬 Ask me about: Generative AI, AI infrastructure, Computer Vision.
Building the turn-geometry layer of voice agents: endpointing, turn-taking, latency and conversation KPIs. Research-backed, browser-runnable instruments — open one, no setup:
| Instrument | What you see | Open |
|---|---|---|
| Voice Agents Playground | Silero VAD endpointing live on your mic — threshold/hangover tuning in real time | ▶ |
| Word Timing Explorer | Whisper word-level timestamps in-browser (barge-in & alignment substrate) | ▶ |
| Latency Gap Simulator | Where the milliseconds go in a voice turn — waterfall, presets robust→S2S | ▶ |
| Voice Metrics Dashboard | Paste a transcript → KPIs: gaps p95, interruptions, containment | ▶ |
| Jev Mini Bench | The tiny decision-model wave, in your pocket | ▶ |
Upstream contributions: livekit/agents#7735 (Silero VAD
threshold-pair fix) · transformers.js#1796
(getAvailableDevices() API).
All products ship code on GitHub + live on Hugging Face. This work is part of The Learning Curve — hands-on AI projects & courses (Substack).
I maintain several repositories focusing on the YOLO ecosystem, adversarial attacks, and edge deployment.
| Project | Description | Stack |
|---|---|---|
| YOLO Adversarial Fullstack | Full-stack application demonstrating adversarial attacks on object detection models. | Python YOLO Web |
| YOLO Active Learning | Framework for implementing active learning loops to optimize dataset annotation. | PyTorch Active Learning |
| YOLO Minimal Inference | A lightweight, stripped-down inference engine for YOLO models designed for speed. | Python ONNX |
| YOLO Trainer Template | A structured boilerplate for training custom YOLO models with best practices. | Ultralytics Python |
| Ultralytics 16U | Specialized implementation utilizing Ultralytics 16U architecture. | Computer Vision |
| Project | Description | Stack |
|---|---|---|
| Gemini Thumbnail Gen | Automating thumbnail creation using Google's Gemini multimodal capabilities. | GenAI LLMs |
| Camera Tracking RPi | Real-time object tracking system optimized for Raspberry Pi hardware. | Embedded OpenCV |
| TLC Agents Training | Training environment for reinforcement learning agents. | RL Agents |
My academic work focuses on industrial applications of Deep Learning and defect detection.
- Defect Detection in Complex Geometries
- Exploring deep fully convolutional neural networks for surface defect detection in complex geometries.
- Read Paper (Springer)
- Trustworthiness in Neural Networks
- Trustworthiness Score for Echo State Networks by Analysis of the Reservoir Dynamics (ESANN 2024).
- Read Paper (ESANN)
- Synthetic Data Generation
- Generating Realistic 3D Surface Defects for Training AI-Based Industrial Inspection Systems.
- Read Paper (Nature Portfolio)
📫 info@amdgarcia.com | 🌐 LinkedIn