Self-Hosted AI Portfolio

This repository contains a fully open-source portfolio with an integrated, self-hosted AI assistant.

The goal of this project is not to showcase AI hype, but to demonstrate pragmatic system design, clear architectural boundaries, and end-to-end ownership of a real application.

---

What this is

A personal portfolio website with an AI assistant that can answer questions about my background, experience, and projects.

The assistant is retrieval-augmented, intentionally constrained, and designed to behave predictably rather than creatively.

In addition, AI is used selectively in supporting roles (e.g. request moderation), where it provides clear, measurable value.

---

Why this exists

This project was built to demonstrate:

• Ownership of the full stack (frontend, backend, infrastructure)
• Thoughtful use of AI under real constraints
• Clear separation between policy, control flow, and model output
• Defensive system design with explicit guardrails
• Systems that fail gracefully and transparently

Everything here is designed to be understandable, reproducible, and adaptable.

---

What makes it interesting

• Self-hosted LLM (no third-party AI APIs)
• Retrieval-Augmented Generation (RAG) over curated knowledge
• Explicit moderation and guardrails
• AI-assisted contact form moderation (spam, abuse, prompt-injection)
• No server-side chat history storage
• Designed for limited hardware (runs on a Raspberry Pi)

AI is treated as a component, not an authority.

---

High-level architecture

Browser
  ↓
Frontend (Next.js)
  ↓
Spring Boot API
  ├─ Request moderation & routing
  ├─ Contact form moderation
  ├─ Retrieval (PostgreSQL + pgvector)
  └─ Prompt assembly
  ↓
Ollama (self-hosted LLM)

---

Technology stack

• Backend: Kotlin / Java, Spring Boot
• Database: PostgreSQL with pgvector
• Frontend: Next.js
• LLM runtime: Ollama
• Deployment: Docker Compose

---

Design constraints

This project intentionally operates under the following constraints:

• Limited compute and memory
• No external AI services
• Predictable behavior over novelty
• Minimal data retention
• Clear, observable failure modes

These constraints drive most architectural decisions.

---

Getting started

Clone the repository:

git clone https://github.com/SebastianoFazzino/portfolio.git
cd portfolio

Start the system:

docker compose up

Refer to the individual service directories for configuration details.

---

Open source

This project is fully open source and licensed under MIT.

You are welcome to:

• Explore the code
• Fork the repository
• Adapt it for your own portfolio or projects

---

License

MIT License.