$ en/home

Home

Technical implementation, structured documentation, automation, and applied AI — the portfolio and technical journal of Fabien Cappelli.

Hello, I’m Fabien Cappelli.

Cyber-gutter-punk. Former palaeographer. PhD in Linguistics. Trained in mathematics, data science, and AI engineering.

I design and implement technical systems built to survive contact with reality: SaaS integrations, APIs, data models, structured documentation, and AI-assisted automation.

My work often begins with a need that is still loosely defined: understanding its business and regulatory constraints, then turning it into workflows, tools, and knowledge that people can actually use.

I am less interested in flashy demos than in systems that are explainable, maintainable, and useful to the people who will rely on them.

What I’m working on

  • building Robie, a local voice assistant on Raspberry Pi combining speech recognition, a language model, and text-to-speech
  • designing and building a MkDocs documentation ecosystem from the ground up, replacing a static PDF guide and supporting both human readers and AI-assisted retrieval
  • building tools that use AI to automate concrete tasks: asking questions about a YouTube video using its transcript and a local model, or turning a Jira export into a narrated release video

Traceability Explorer

A Python tool that starts from a purchase order, recursively reconstructs a traceability chain distributed across multiple API environments, generates an explainable Mermaid graph, and flags data inconsistencies before validation.

Python REST API JSON Mermaid Graph Traversal Data Validation

Designing a Multi-Channel Documentation Ecosystem

A look back at the creation of an intelligent documentation pipeline that transforms a MkDocs knowledge base into a multi-channel publishing system compatible with chatbots and AI: rendered HTML extraction, two-pass synchronisation, automatic link rewriting, intelligent Mermaid diagram handling, and the gradual emergence of a true documentation graph that AI agents can use.

MkDocs Material for MkDocs Markdown Python Git GitLab CI/CD

Robie

Voice assistant on Raspberry Pi

Python Vosk TTS

Release Video Generator

Python pipeline that transforms a Jira HTML export into a narrated release video, with thematic script generation, text-to-speech synthesis, and automated video rendering.

Python OpenAI API BeautifulSoup FFmpeg Pydub

Semantic Image Segmentation (OpenClassrooms Projects 8 & 9)

Semantic image segmentation applied to intelligent vision systems using urban scenes from the Cityscapes dataset.

Python Keras Segmentation Models FastAPI Streamlit Docker

Tweet Sentiment Prediction (OpenClassrooms Project 7)

Sentiment analysis on 1.6M tweets with comparison of NLP models (TF-IDF, CNN, LSTM, DistilBERT), full MLOps workflow, cloud API, and CI/CD deployment.

Python Pandas Scikit-learn Keras Hugging Face Transformers DistilBERT

Latest posts

Now We're Getting Serious...

Moving into V1.2: intent detection, conversational architecture, and the first real foundations of the agent

Chatbot Integration v1

A look back at the creation of an intelligent documentation pipeline turning a MkDocs knowledge base into a multi-channel publication system compatible with chatbots and AI: rendered HTML extraction, two-pass synchronization, automatic link rewriting, smart Mermaid diagram handling, and the gradual emergence of a true documentation graph usable by AI agents.