Skip to content

Rafik Mammeri

I take AI agents from whiteboard to production — and make them part of how the company works.

Technical lead for AI agents at Boulanger, one of France's largest electronics retailers (€4B). Four systems in production across three channels: customer web & mobile, voice, and the tools employees use every day. Before that, seven years shipping machine learning in regulated banking. PhD in mathematics.

~2,000conversations / day 75%resolved end-to-end 78%satisfaction, 10,000+ ratings 6 mofrom zero to production

Selected work

  • Customer AI assistant — web & mobile


    Replaced a static decision-tree bot with a multi-agent assistant covering the full customer journey. Now a permanent customer channel at a €4B retailer.

    ~2,000conversations / day 75%resolved end-to-end

    Case study

  • Self-BI — the warehouse, in plain language


    Business teams query Snowflake conversationally — no analyst in the loop, no Snowflake account needed, no raw SQL ever shown.

    MCPGemini Enterprise

    Case study

  • Vox — voice callbot API


    The assistant rebuilt for the phone, where latency is silence: token streaming so speech starts before the answer ends, zero LLM calls where a rule suffices.

    SSE streamingHexagonal

    Case study

  • Internal agents — Google Chat


    Business-owned agents on the tool every employee already has open — starting with the company's official HR assistant.

    RAGAgent Engine

    Case study

How I work

Architecture before prompts

The biggest latency and cost wins in LLM systems are structural. The assistant answers in one or two model calls per message not because of clever prompting, but because routing and handoffs are native to the graph.

Design for the model being wrong

Four independent security layers stand between an agent and the data warehouse. Even a hallucinated query lands on a read-only role scoped to its use case — and gets rejected.

The business owns the product

The HR assistant is owned by HR — content, scope, response validation. My job is making sure the technical layer never gives them a reason to doubt it.

Production is a loop, not a launch

Versioned prompts, LLM-judge scoring against live traffic, and continuous releases since day one. Quality in production isn't achieved — it's maintained.

Writing

All articles →

Experience

  • 2025 — present Senior AI Engineer & technical lead for AI agents, Boulanger
  • 2024 — 2025 AI / DevOps Engineer (Freelance), ITS Group
  • 2021 — 2024 Senior Data Scientist, BNP Paribas Datalab
  • 2018 — 2021 Data Scientist / Risk Manager, BNP Paribas Risk

Full experience & education →

Skills

LLM & Agents

LangGraphLangChainGoogle ADKMCPRAGAzure OpenAIGeminiLangfuse

Engineering & MLOps

PythonFastAPIKubernetesDockerHexagonal architectureAsync / SSECI/CD

Cloud & Data

GCP / Vertex AIAzureSnowflakeMongoDBTerraformAPI gateways