Ramzi B.

Development of a conversational AI assistant capable of answering team questions by leveraging the company's internal documentation.

The chatbot uses a RAG (Retrieval-Augmented Generation) architecture to provide sourced and accurate answers, citing the original documents.

Integration of data sources: internal PDFs, Notion base, Google Drive documents.

The documents are vectorized and indexed in a vector database for fast semantic search. Each answer includes the source and the exact passage used, eliminating hallucinations.

Deployed on secure infrastructure with data hosted in France. No data is transmitted through third-party servers. Interface accessible via a web widget integrated into the company's internal site.

Result: 60% reduction in recurring questions addressed to managers and the internal support team. Average response time: 8 seconds versus several hours of waiting.

Setup of a complete production infrastructure for self-hosted n8n, intended to replace a limited and expensive n8n Cloud instance.

Full containerization with Docker and orchestration of services (n8n, PostgreSQL database, Nginx reverse proxy).

Deployment on a secure cloud server.

Network and firewall configuration for minimal exposure.

Daily automated backups with 30-day retention and the ability to restore in under 15 minutes.

Real-time monitoring via Datadog: alerts for failed workflows, CPU/RAM/disk monitoring, and a performance dashboard accessible to the client.

Infrastructure fully defined as code with Terraform: reproducible, versioned, and auditable environment.

Complete technical documentation delivered to enable the client's team to understand and maintain the infrastructure.

Result: 99.9% uptime. Incident detection time under 2 minutes. Monthly cost of €50 versus €350/month on n8n Cloud for equivalent usage, saving €3,600/year.

Design of an AI agent that automatically processes incoming leads for a real estate agency. Each request received (web form, email) is analyzed by the AI: extraction of the need, budget, desired location, and urgency level.

The agent assigns a qualification score based on client-defined criteria, then automatically routes the lead to the correct advisor on Slack with a complete summary sheet. Unqualified leads receive a personalized automatic response with next steps.

n8n workflow with webhooks for real-time reception, OpenAI API for analysis and scoring, Slack API for routing, and Gmail for automatic responses.

Complete documentation delivered for team autonomy.

Result: automatic processing of 100% of leads in less than 2 minutes, compared to 4 to 6 hours for manual processing. No leads lost or forgotten.