Axel Stoltz

AI agent (LangGraph) in the form of a chatbot capable of generating complex DSL scripts on demand and navigating language documentation, without support or prior training. Knowledge base via RAG pipeline, advanced optimization on lightweight models.

Result: learning curve reduced by approximately 80%, autonomous client deployment.

Sole technical profile on the project, in direct contact with operational teams. The objective: to automate the ingestion and structuring of 50,000+ product references that were manually accumulated from fragmented supplier sources.

I built extraction pipelines combining NLP, classification, and scraping on heterogeneous web pages and technical documents. This project is directly at the origin of SonarFetch, which I subsequently founded to industrialize these workflows on a larger scale.

SonarFetch is a B2B AI SaaS that I founded to solve a concrete problem: distributors and buyers of electronic components spend a huge amount of time manually searching for product data from dozens of different supplier sources, all in incompatible formats.

The platform automates this collection, normalizes the data with LLMs (Mistral), and makes it directly usable in business workflows. But building the pipeline was not enough: I worked just as much on the interface to make it usable by anyone in an operational team, without training, without documentation. Because a tool that people don't use is worthless.

Stack: FastAPI, Celery, PostgreSQL, Typesense, React, Mistral.

What this changes concretely: tasks that took days are done in minutes, without manual intervention.