Skip to content
FRANCESC CASTELL
— PROJECTS

Tools and systems I've built.

From internal CLIs that save hours every week to self-hosted AI infrastructure. Each project started as a real frustration I wanted to make disappear.

01 Active

Markus

Personal AI assistant with GraphRAG memory and clinical tools.

2026

A conversational AI companion that uses a Neo4j graph database to remember deep context — who you are, how you felt, and the causal chains behind your emotions. Unlike vanilla vector RAG, Markus extracts entities (people, emotions, topics) and links them with semantic relationships. Includes 16 integrated tools: Cognitive Behavioral Therapy exercises, SMART goal coaching, journaling, crisis intervention, and an interactive mind graph. Fully self-contained in Docker with zero SaaS dependencies.

  • Neo4j graph memory with entity extraction and causal relationship mapping.
  • 16 autonomous tools — CBT exercises, SMART goals, journaling, crisis intervention.
  • Streaming chat interface with LLM-driven function calling and mind graph visualisation.
Python Neo4j GraphRAG FastAPI Docker LLM AI
02 Active

Homer CLI

One tool to unify the Terraform and Packer lifecycle.

2025

A wrapper over Terraform and Packer that unifies and simplifies infrastructure lifecycle management. Flexible syntax and intuitive commands (plan, apply, build, destroy) ensure consistency across environments. Includes homer unlock to safely release Terraform state locks without leaving the terminal.

  • Single command interface for plan/apply/build/destroy across multiple environments.
  • Safe state-lock release with `homer unlock`.
  • Started as a 3-line alias, grew into a 400+ line CLI adopted internally.
Python Bash Terraform Packer CLI
03 Shipped

AWS High Availability Architecture

Resilient multi-AZ stack with SSM and CloudWatch.

2025

Terraform codebase that provisions a complete high-availability AWS stack — VPC across multiple AZs, ALB in public subnets, EC2 Auto Scaling Group in private subnets with NAT Gateways, CloudWatch Agent for custom metrics, and Systems Manager for secure access without SSH. Built for production workloads that demand resilience.

  • Multi-AZ deployment with private subnets, NAT Gateways, and ALB for fault tolerance.
  • CloudWatch Agent collecting CPU, memory, and disk metrics at 60s intervals.
  • SSM Session Manager for secure, SSH-less instance access.
AWS Terraform ALB ASG CloudWatch SSM HA
04 Shipped

Fermy Academy

Django web platform for tutoring and academic support.

2023

A full-stack web application built with Django and Bootstrap for managing tutoring courses and academic support services. Includes production deployment checks and SQLite-backed data persistence.

Python Django Bootstrap Full-stack
— Also built

Terraform · Packer Wrapper

Shipped

Interactive target selection for Terraform deploys.

Private LLMOps · Self-hosted RAG

Active

An on-premise RAG ecosystem with zero cloud dependency.

2025

AWS Cloud-Native Web App

Shipped

Highly scalable and secure web application architecture.

Portfolio

Active

This very site — built with Astro 5 and Tailwind v4.