Matthieu Pesesse

Projects

Software I designed and built — client work and lab projects. The first set is live and clickable; the second runs on my own machine.

Built & live

Online right now

Client sites and lab projects with a public URL — including the full dev stack, the automations that keep them running, and the chatbot powering each one.

Abbys Consult — screenshot

Abbys Consult

2026

Brand site for an independent Belgian consulting firm

Full brand site for Abbys Consult, the independent advisory practice of Christophe Vandemoortele — strategy, organisational transformation and performance optimisation across the Benelux.

Stack

HTML/CSS/JS static siteNode.jsExpressnginxSSE streamingNVIDIA NIM

Automations

  • Express chatbot backend kept hot under PM2 so the primary model answers with a sub-second warm time-to-first-token.

Chatbot

Llama 4 Maverick 17B → Llama 3.1 70B · NVIDIA NIM

Conversational assistant on an Express backend, served over SSE streaming with a non-stream fallback. Multi-provider chain: NVIDIA NIM running Llama 4 Maverick 17B as primary (kept hot, ~0.5s warm TTFT) with automatic fallback to Llama 3.1 70B. Temperature 0.5, 1024 max tokens.

Visit live
Spotify Showcase — screenshot

Spotify Showcase

2026

Self-built showcase for my own Spotify catalogue

An in-page player and showcase for my own electronic music catalogue — trance, techno, dreampop, orchestral — built on React + Vite, fed by a Python scraper.

Stack

ReactVitePythonScraplingBlotato APIClaude Haiku

Automations

  • Daily "Music" Facebook post published at 20:00 Brussels.
  • Daily trance-only "Tranceportation" Facebook post published at 21:00 Brussels.
  • Friday 08:30 full refresh: scrapes Spotify, patches the bundle, redeploys, and appends new trance releases — captions generated by Claude Haiku, posts published through Blotato.
Visit live
Matthieu Pesesse — screenshot

Matthieu Pesesse

2026

This very site — built and shipped end to end

The site you are reading. A trilingual Next.js 15 App Router portfolio with a daily auto-blog and a full structured-data layer for AI discoverability.

Stack

Next.js 15 App RouterReactTailwindPostgreSQLbunPM2nginx

Automations

  • Daily auto-blog at 08:00 Paris — generates a trilingual AI-news article, prerenders it, pings IndexNow and auto-shares on LinkedIn.
  • Content sync + standalone rebuild + PM2 reload chained automatically after each publish.
  • Weekly newsletter recap every Monday at 07:00.

Chatbot

Llama 4 Maverick → Llama 3.1 70B → Kimi K2 · NVIDIA NIM + OpenRouter

Conversational agent streamed over SSE. Multi-provider chain: NVIDIA NIM Llama 4 Maverick 17B → Llama 3.1 70B → OpenRouter Kimi K2, so a model is always answering even under provider load.

Visit live

Running locally

On my own machine

AI tooling and automation that runs on my MacBook — no public URL, shown as screenshots. Same detail: stack, automations, and the model behind each agent.

Neurolinks Hub — screenshot

Neurolinks Hub

2026

A central command centre for every project and agent on the machine

A self-hosted command centre: a dashboard, a registry of skills and tools, an MCP server, a port manager and an activity monitor — the shared backbone every other local project plugs into.

Stack

PythonVanilla JSMCP serverSQLite registry

Automations

  • Auto-update sweep every 6 hours — rescans and refreshes the skills and tools registry.
  • Security scanner gates every new skill before it can enter the registry.
  • MCP server exposes the registry, shared config and shared files to every Claude Code agent on the machine.
Agents Neurolinks Pro — screenshot

Agents Neurolinks Pro

2026

A Telegram-driven multi-agent operating system

A multi-agent OS driven from Telegram: a CEO orchestrator delegating to Architect, Coder, Tester, Reviewer, Communicator and Memory agents, with a brutalist Angular 19 telemetry dashboard.

Stack

TypeScriptNode.js 22Angular 19 SignalsPostgreSQL + pgvectorRedis + BullMQDrizzle ORMExpress + SSENVIDIA NIM

Automations

  • BullMQ on Redis queues and runs agent jobs; pub/sub streams live state to the dashboard.
  • pgvector HNSW index gives every agent long-term semantic memory.
  • Telegram bot auto-relaunches on conflict so the command channel stays up.

Chatbot

Telegram-driven · NVIDIA NIM tiered (Nemotron, Qwen3-Coder 480B, Llama 3.3 70B)

The whole system is conversational — driven from Telegram on an allowlist. A CEO orchestrator routes each request to specialist agents (Architect, Coder, Tester, Reviewer, Communicator, Memory). Models are tiered on NVIDIA NIM: Nemotron-3 super/nano for reasoning, Qwen3-Coder 480B for code, Llama 3.3 70B general, and nv-embedqa-e5-v5 for retrieval.

LinkedIn SSI Growth — screenshot

LinkedIn SSI Growth

2026

An approval-mode LinkedIn growth engine

A personal LinkedIn growth engine that stays in approval mode: it tracks the Social Selling Index, drafts French hooks, and queues posts and engagement for review before anything goes out.

Stack

TypeScriptPlaywrightExpressClaude CLIBlotato APISQLite

Automations

  • Social Selling Index tracked over time from a Playwright persistent profile.
  • French post hooks drafted by Claude CLI, held for manual approval.
  • Approved posts published through the Blotato API; engagement and outreach queued the same way.

Ready to create something amazing together?

Let's discuss how I can help bring your vision to life through strategic design that delivers tangible results for your business.

Get in touch
Projects | Matthieu Pesesse | Matthieu Pesesse