Alejandro Golfe — AI Engineer

AI Engineer with 5 years of hands-on experience in generative AI, computer vision, and LLM-based systems. PhD candidate at UPV with published research in medical imaging and production deployments spanning voice agents, inference pipelines, and cloud infrastructure.

📧 alejandrogolfe@gmail.com · 📍 Valencia, Spain

Featured Projects

🔬 BlastDiffusion — Synthetic Embryo Generation

Code · Publication

Latent Diffusion Model that generates synthetic oocyte images conditioned on developmental outcomes, addressing critical data scarcity in IVF datasets. Outperforms GAN baselines in image quality and captures key morphological differences linked to blastocyst progression.

BlastDiffusion Methodology

PyTorch Diffusion Models VAE Medical Imaging Data Augmentation

🎙️ Wellbeing Conversational Voice Agent — 2026

Production real-time voice agent for hospitalised patient wellbeing with guided conversation flow, persistent memory, and real-time state evaluation.

Reduced end-to-end latency 30% using Deepgram Nova-2 (STT) and Cartesia Sonic (TTS), validated with production logs
Built observability pipeline: LangSmith tracing, latency profiling per component, and conversation quality evaluation

Python LiveKit OpenAI API Cartesia Deepgram Firebase LangSmith

🎙️ Hospital Survey Voice Agent — 2025

Production voice agent for automated patient satisfaction surveys with stateful multi-mode conversation flow (general, deepdive, missing KPI recovery). LLM calls kept strictly analytical with Python-driven logic.

Real-time Firebase persistence with async task management and fault-tolerant state handling

Python LiveKit OpenAI API AWS Polly Firebase

🔍 CLAV — WSI Retrieval with Foundation Models

Code · Publication

Content-Based Image Retrieval method that leverages foundation models pre-trained on histopathology data without task-specific training. CLAV condenses feature vectors into a unique representative vector per Whole Slide Image, reducing memory bank size while outperforming state-of-the-art retrieval methods.

CLAV Methodology

PyTorch Foundation Models CBIR Histopathology WSI

🧬 ProGleason-GAN — Prostate Cancer Patch Synthesis

Code · Publication

Conditional Progressive Growing GAN for synthesising histopathological patches conditioned on Gleason grade. Validated by expert pathologists and demonstrated significant classification improvement on SiCAPv2.

ProGleason-GAN Methodology

PyTorch GANs Progressive Growing Medical Imaging Data Augmentation

Technical Skills

Generative AI & Deep Learning — PyTorch, TensorFlow, GANs, Diffusion Models, Transformers, Hugging Face
Computer Vision — OpenCV, torchvision, YOLO, CBIR, medical image analysis
LLM & Agents — LangChain, LangGraph, OpenAI API, LiveKit Agents, LangSmith
MLOps & Deployment — FastAPI, Docker, MLflow, Weights & Biases, ONNX, GitHub Actions
Cloud & Infrastructure — AWS (S3, Lambda, ECR, event-driven pipelines), Terraform
Evaluation — LangSmith tracing, custom evaluators, latency profiling, RAGAS

Publications

CLAV: Clustering Latent Vector Aggregation for WSI Retrieval Leveraging Foundation Models (2025). Knowledge-Based Systems
Enhancing Image Retrieval Performance with Generative Models in Siamese Networks (2025). Journal of Biomedical and Health Informatics
ProGleason-GAN: Conditional Progressive Growing GAN for Prostatic Cancer GG Patch Synthesis (2023). Computer Methods and Programs in Biomedicine

Education

PhD in Health and Wellness Technologies (AI Focus) — UPV (2022 – Present)
Master’s in Artificial Intelligence — VIU (2021 – 2022)
Master’s in Industrial Engineering — UPV (2018 – 2021)
Bachelor’s in Industrial Technologies Engineering — UPV (2014 – 2018)