Antoine D.
✓ Vetted CrafterPrincipal Software Engineer – Integrations & New Features
I connect systems and build APIs — integrations that scale and features that ship.
About
I build integrations and product features. I have connected 15+ products to external services: payment gateways, CRMs, marketplaces. I specialise in REST APIs, webhooks, and event-driven architectures. Cursor helps me scaffold integration code and debug API issues. I work with TypeScript, Node.js, and PostgreSQL.
My focus is on making products connect: OAuth flows, webhook handlers, sync jobs, and API design. I have built integration platforms that connect e-commerce to 12+ services, and I care about reliability — retries, idempotency, and error handling. Good integrations are invisible until they break; I make sure they do not.
AI Expertise
Notable Projects
Legal Document Summarization with Fine-Tuned Mistral
Fine-tuned Mistral 7B using QLoRA on a dataset of 50,000 French legal document/summary pairs for a law firm. Used Hugging Face PEFT and TRL for RLHF alignment to summary quality preferences. Deployed with vLLM for 8x throughput improvement over naive inference.
✓ Outperforms GPT-4 on French legal summarization (measured by lawyer quality ratings) at 95% lower cost per summary; processes 1,200 documents per hour.
Enterprise Knowledge Agent with Mistral Function Calling
Built an enterprise knowledge agent using Mistral's function calling API for a French industrial company. The agent has access to tools for querying an internal knowledge base (LlamaIndex + Weaviate), fetching from SharePoint, and querying structured databases. Runs entirely on EU infrastructure for GDPR compliance.
✓ Knowledge retrieval accuracy of 89% on internal evaluation set; passed GDPR data processing audit; adopted by 1,200 employees within 2 months of launch.
French Biomedical NLP Benchmark
Led an open-source project to create and publish a comprehensive French biomedical NLP evaluation suite covering NER, relation extraction, and clinical QA. Evaluated 12 models including CamemBERT, multilingual models, and fine-tuned Mistral variants.
✓ Benchmark adopted by 6 French research labs; fine-tuned Mistral model outperformed all baselines and was open-sourced with 800+ Hugging Face downloads in first month.
Work Experience
Principal NLP Engineer
Doctrine (Legal AI)
2020 – Present
Lead NLP research and engineering for an AI-powered French legal research platform. Own the model fine-tuning roadmap, the evaluation framework, and the RAG architecture for case law search over 15 million French legal documents.
Research Scientist, NLP
INRIA
2014 – 2020
Published NLP research on machine reading comprehension, neural summarization, and cross-lingual transfer. Contributed to the FLUE (French Language Understanding Evaluation) benchmark.
Education & Certifications
Ph.D. Natural Language Processing
Sorbonne Universite · 2014
Interested in working with Antoine?
Tell us about your project and we'll facilitate an introduction.