Skip to content
Ton Llop

InfrastructureDevOpsCybersecurityLocal AI

Building systems that keep working when things go wrong.

I’m Ton Llop, a final-year engineering student in Tarragona. Most of what I build is backend code and the infrastructure it runs on, and lately I’ve been moving toward DevOps, security and running AI models locally.

My projects follow a system from setup to failure and back: provisioning it, automating it, securing it, watching it and restoring it. Along the way that has meant Debian servers, containerized workers, a scientific platform and local AI runtimes.

Now / Final-year thesis on local AI · open to DevOps and infrastructure internships

Based in Tarragona, Spain · English / Spanish / Catalan

Focus map

  • Infrastructure & DevOpsprovision · automate · observe · recover
  • Cybersecurityharden · verify · audit
  • Local AIlocal inference · privacy · control

Foundation / Backend engineering · Distributed systems

I build and operate reliable systems, explore how to secure them, and research how AI workloads can run locally.

Every system fails eventually. The craft is in what happens next.

01Selected work

Scientific software

Lipopotamo

A modular scientific platform for processing and managing NMR data. It handles everything from uploaded batches and single-sample runs to plots, outputs and reports.

I work on execution management across batches, samples and instances: selective runs of scientific software, asynchronous processing with specialized workers, orchestration with execution history and partial re-runs, merge strategies for scientific results, and authentication and access control across containerized services.

  • Batch, sample & execution management
  • Asynchronous workers & orchestration
  • Execution history & partial re-runs
  • Python
  • FastAPI
  • MongoDB
  • React
  • TypeScript
  • Keycloak
  • Docker Compose
Internal project

Private repository

Local AI infrastructure

JaullLocal LLM qualification

Checks whether a local LLM will really work on a given machine, starting from the hardware it actually has and ending with benchmarks measured on it.

Jaull analyses local hardware, selects candidate model configurations, resolves and verifies the exact artifact that would run, validates runtime and backend readiness, executes the model locally and compares its predicted requirements against real benchmark measurements. Workload-aware qualification and reproducible deployment are the direction it is heading, not what it does today.

  • Hardware-aware model qualification
  • Artifact verification & provenance
  • Real execution & benchmark evidence
  • Prediction vs observed performance
  • Python
  • Typer
  • Textual
  • llama.cpp
  • GGUF
  • Hugging Face
Work in progressjaull.techView on GitHub
Security tooling

Ninjadorks

A search-automation toolkit for authorized security research. It builds advanced queries, saves the results in a structured format and helps you analyse whatever turns up.

Automates Google and DuckDuckGo searches with advanced dork construction, produces structured JSON and HTML reports, retrieves selected files and applies regular-expression and optional model-assisted analysis. Intended for authorized research, education and defensive analysis in controlled environments.

  • Advanced dork construction
  • Structured JSON & HTML output
  • Regex & model-assisted analysis
  • Python
  • Selenium
  • HTML
  • JSON
  • Regex
Infrastructure & server administration

Server to ClusterDebian → Kubernetes, run as code

Two stages of the same infrastructure work. First a Debian server operated as code, then the same discipline scaled up to containers with Docker Compose, Kubernetes, Terraform and a full observability stack.

The server stage can be rebuilt from its repository: idempotent provisioning, hardened key-only SSH, systemd services and timers as code, and restore-tested backups with disaster-recovery runbooks. The platform stage runs an Nginx frontend, a Flask API and Redis on Kubernetes, segments traffic with NetworkPolicies, provisions the same stack with Terraform and watches it with Prometheus, Alertmanager and Grafana.

  • Idempotent provisioning & systemd as code
  • Restore-tested backups & recovery runbooks
  • Kubernetes, NetworkPolicies & Terraform
  • Prometheus, Alertmanager & Grafana
  • Debian
  • Bash
  • systemd
  • Docker Compose
  • Kubernetes
  • Terraform
  • Prometheus
  • Grafana
Mobile application

Beer Fantasy

A social mobile application for tracking drinks, competing with friends and turning nights out into fantasy-style leagues and challenges.

Designed and built end to end: the Flutter app, authentication, the PostgreSQL data model on Supabase with Row Level Security and RPC functions, the social and league logic, n8n automation workflows and the full Android release on Google Play.

  • Leagues, leaderboards & weekly competition
  • Achievements & statistics
  • Moderation & responsible-use limits
  • Flutter
  • Dart
  • Supabase
  • PostgreSQL
  • Row Level Security
  • n8n
  • Android

02More work

  • DD-tmux

    Developer tooling

    My AI coding agents as a Darkest Dungeon party. Every tmux session becomes a torch-lit room and every agent a hero that fights while it works, sleeps when it's done and raises an alert when it needs me. It detects 14 agents, opens a live terminal for each one and keeps a chronicle of every turn in Postgres.

    Python · tmux · PostgreSQL

  • Homelab

    Self-hosting · learning in progress

    A small homelab I'm building to learn: an old desktop running Proxmox, with Pi-hole for DNS and ad blocking, a Homarr dashboard and a power monitor I wrote around a smart plug. Still early days; backups, a reverse proxy and Ansible are next.

    Proxmox · Docker · Pi-hole · Python

  • Scalable & Elastic Ticket Service

    Distributed systems

    Queue-based ticket processing with RabbitMQ, stateless workers and PostgreSQL on AWS ECS Fargate. Processing is idempotent and at-least-once, with retries, dead-letter queues and autoscaling based on the queue backlog. I benchmarked it for throughput and latency.

    RabbitMQ · PostgreSQL · AWS Fargate · Docker

  • Superscalar simulation & branch prediction

    Computer architecture

    Simulation of a superscalar processor and analysis of an Alloyed branch predictor, modeled in C.

  • Graph centrality at scale

    Algorithms & data

    Vertex-importance and centrality measures computed over large GraphML graphs, implemented in Java.

    Java · GraphML

All repositories on GitHub

03About

Portrait of Ton Llop

Ton Llop

Infra engineer · Tarragona

Quirks

  • +Calm under pressure · former lifeguard
  • +Homelab tinkerer · Proxmox runs at home
  • +Both ends of the stack · from Verilog to Kubernetes
  • −Just one more container · the homelab keeps growing

Speaks Catalan · Spanish · English · a little German

I’m Ton. I like understanding a system all the way down, from the API someone calls to the machine it ends up running on, and I like it even more when it keeps working after something breaks.

I spent several summers as a lifeguard on the beaches and pools of Tarragona. Most days were quiet, and the work was staying ready for the day that wasn’t. I try to run infrastructure with the same habit, so the monitoring, the backups and a recovery plan are in place before anything breaks.

At home that turns into a homelab, an old desktop running Proxmox with my own DNS, a dashboard and a power monitor, which mostly exists so I can break things safely. With friends it turned into Beer Fantasy, an app for our fantasy league that ended up on Google Play.

I’ve also spent time at the other end of the stack, writing low-level C and building a MIPS CPU in Verilog. Having seen both ends, I find it easier to reason about a system as a whole.

Now

  • Final year of Computer Science and Engineering at URV
  • Thesis on choosing and running AI models locally
  • Building Jaull and working on Lipopotamo

Open to

  • Internships · junior roles
  • Infrastructure · DevOps · backend
  • Cybersecurity · distributed systems
  • Local AI · scientific software

04Technical focus

  • Infrastructure & cloud

    DockerDocker ComposeKubernetesTerraformAWSPrometheusGrafanaLinux

  • Security & DevSecOps

    System HardeningBackup & RecoveryNetwork segmentationNetworkPoliciesAuthorized Security ResearchRegex analysis

  • Local AI

    llama.cppLocal model inferenceModel & runtime managementArtifact verification & benchmarkingHardware-aware execution

  • Backend & data

    PythonFastAPIPostgreSQLMongoDBSupabaseRedisRabbitMQREST APIsWebSocketsAsynchronous processing

  • Frontend & mobile

    ReactNext.jsTypeScriptTailwind CSSViteFlutterDartAndroid

  • Systems

    CJavaVerilogDistributed systemsQueue-based architecturesWorker architecturesComputer architecture

A backup never restored is a prayer, not a plan.

05Contact

Let’s connect.

If you’re hiring for an internship or a junior role in infrastructure, DevOps or backend, or you just want to talk about one of these projects, write to me.