About Anas Rabhi — Freelance AI Engineer & Data Scientist¶

My name is Anas Rabhi. I am a Data Scientist and freelance AI engineer, based in Toulouse. My work: helping companies move from "we'd like to do something with AI" to "it's running in production and generating value".

No grand talk about digital transformation. Code that works, systems that hold up under load, and results you can measure.

Why AI, and why freelance¶

I started with classical Data Science — predictive models, time series, NLP. Then LLMs changed what was actually feasible. They didn't replace scientific rigour, they just expanded the scope of what you can build in a matter of weeks.

Freelancing is a deliberate choice. I want to stay close to real problems, not locked in a silo. On every engagement, I am involved from scoping through deployment — and often beyond, for production monitoring.

What I work on¶

My expertise revolves around four areas:

• RAG (Retrieval-Augmented Generation) — hybrid BM25 + vector retrieval, reranking, chunking, quality evaluation. This is the core of most of my projects. I write about it regularly on the blog.
• AI Agents — tool orchestration, multi-step workflows, production guardrails. The classic mistake I avoid: aiming for a generalist agent from day one.
• NLP — classification, information extraction, summarisation, semantic search on business data.
• LLMOps — monitoring, cost management, security, deployment and maintenance of models in real conditions.

On complex projects, I also build exploratory notebooks to validate hypotheses before committing to a full solution.

My two businesses¶

Tensoria is my AI consulting firm. I work there on a time-and-materials or fixed-price basis for scoping, prototyping, RAG audits, production deployment and data team coaching. Primarily in Toulouse and remote. The first conversation is always free.

Heeya is a product born from a recurring client pattern: deploy a personalised RAG chatbot on any website, trained on its own content, in a few clicks. A turnkey solution for companies that want an AI assistant without building a data team.

How I work — my engagement method¶

Every project follows the same structure, regardless of size.

Scoping (1–2 weeks). I always start by understanding the business need before choosing a technology. What problem are we solving? For whom? Which KPIs define success? At this stage, I often push teams to drop unnecessary features — a clear scope is half the work.

POC on real data (2–4 weeks). No mockups, no synthetic data. A POC must run on a representative subset of the client's actual data. That is the only way to surface real problems — document quality, edge cases, latency.

Production deployment. Integration with existing systems, access management, monitoring, guardrails. This is where the majority of AI projects fail — not on the model, but on infrastructure and robustness.

Monitoring and iterations. An AI system degrades if left unwatched. I put technical metrics (recall, hallucination rate, latency) and business metrics (time saved, resolution rate) in place from day one, not as an afterthought.

Measurable results¶

A few outcomes measured on real projects:

• 75% time saved on writing responses to requests for proposals, via an AI agent fed by internal documentation — 300% ROI in the first year.
• 60% time saved on machine-fault diagnosis in a factory, through an AI assistant for operators.
• 2× faster structured information extraction from PDFs, invoices and contracts.
• A sovereign RAG on confidential space data, with zero data leaving the client perimeter (self-hosted LLM, private cloud).

These figures come from real projects, not estimates. I detail them when relevant in articles on the blog.

What I write about¶

I write regularly on applied AI — not generic tutorials, but field notes on what actually works in production.

Recent articles:

• Optimising your RAG: the 8 techniques that really make a difference — HyDE, reranking, contextual retrieval, semantic cache.
• Optimal chunking for your RAG — 8 strategies compared, real benchmarks.
• Training, fine-tuning or RAG: which to choose for your AI? — complete decision tree with cost estimates.

Get in touch¶

If you have an AI project to scope, a RAG architecture to audit, or simply a question about something I've written — get in touch. The first conversation is free and without commitment.

Discuss a project Back to home