Sergio - Full Stack Software Engineer & Applied AI Engineer FULL STACK & APPLIED AI FULL STACKAPPLIED AIFULL STACK

My final-year project explored experimental prostate-tissue classification using the PANDA dataset. I built the data-processing, training and evaluation pipeline around a DenseNet121 convolutional network.

Whole-slide images and masks become labelled patches. Balanced splits, conditional augmentation and focal loss support training. Evaluation includes Cohen's kappa, F1, confusion matrices and ROC curves.

259,044

Patches processed

86.69%

Reported accuracy

0.765

Quadratic kappa

0.8675

Weighted F1

Reported experimental results from my project summary. Not independently reproduced here. Research only, without clinical validation or use in diagnosis.

A technical company needed a useful way to work with approximately 12,000 PDF files. I developed the complete application around a retrieval-augmented AI system.

LangGraph and LangChain coordinate the AI workflow. FastAPI exposes the backend, Next.js provides the interface, and Google Cloud runs the application. The work spans retrieval, application development and cloud deployment.

1. Documents
2. Retrieval
3. AI workflow
4. Application
5. Google Cloud

The archive and answer shown above are conceptual. Internal documents, company identity and unverified performance claims are omitted.

LANGGRAPH / LANGCHAIN / FASTAPI / NEXT.JS / GOOGLE CLOUD

I worked across the complete development process of an enterprise application: understanding the workflow, designing the product, building the frontend and backend, integrating data and delivering the application.

Next.js, FastAPI and PostgreSQL connect the interface to the system. Azure supports deployment and operation. Permissions, change history, monitoring and separate environments are part of the engineering work.

1. Design
2. Frontend
3. API & data
4. Validation
5. Deployment

The project remains anonymous. The interface, scenarios, chart and architecture object are original conceptual illustrations, not internal screenshots or measured results.

NEXT.JS / FASTAPI / POSTGRESQL / AZURE

I help companies put agentic coding into everyday development. Training, reusable resources, skills and playbooks give teams practical ways to work with AI coding tools.

I also deliver in-person vibe coding workshops at universities in Sevilla, with hands-on use of AI coding tools. The focus is on moving from a prompt to a working piece of software.

Harnesses, loops and graphs shape the workflow. Skills, hooks and plugins make it reusable. Claude Code, Codex and Pi belong to this toolkit; the process around them matters as much as the tool itself.