Naji Najari, Ph.D.
I am a Senior AI Engineer
Building multi-agent platforms, RAG pipelines, and LLM-powered products at Brevo.
PhD in unsupervised anomaly detection.
LangGraph
Google ADKRAG
Langfuse
FastAPI
MCP
GCP
AWS
Kubernetes
Palantir
Python
MLFlow
HuggingFace
LinkedIn
GitHub
ScholarAbout
Senior AI Engineer with a PhD in Machine Learning. I help companies ship AI agents, RAG systems, and LLM-powered products to production.
AI Agents
I design multi-agent systems that go beyond chatbots. Agents that call your APIs, query your data, and execute real business logic. Built to run 24/7, not just during a demo.
LangGraph, Google ADK, FastAPI, MCP
RAG Pipelines
I build retrieval systems that give your users accurate answers from your own documents. Hybrid search, reranking, chunking strategies, and proper evaluation to avoid hallucinations.
RAG, Hybrid Search, LangChain
LLM Evaluation
I set up tracing, LLM-as-a-judge scoring, and human review pipelines so you can measure your AI quality and improve it continuously.
Langfuse, LLM-as-a-Judge, HITL
Model Fine-tuning
When a generic model isn't enough, I fine-tune open-source LLMs on your data. Custom classifiers, domain-specific generation, multilingual NLP. From training to serving.
LoRA, PEFT, vLLM, HuggingFace
Experience
Senior AI Engineer
Brevo
- Built multiple production agents within a multi-agent AI platform used by end-users and internal teams: documentation Q&A (RAG), CRM contact management, campaign analytics and recommendation | LangChain, LangGraph, Google ADK
- Exposed Brevo's public APIs as MCP servers, enabling developers to use Brevo tools directly from AI coding assistants (Claude, Cursor) | FastMCP, FastAPI
- Implemented production-grade RAG pipelines for real-time customer query answering | Ragflow, LangChain
- Integrated Langfuse for end-to-end LLM tracing and monitoring; built LLM-as-a-judge evaluation with human-in-the-loop annotation
- Fine-tuned BERT for multilingual spam and fraud detection | HuggingFace Transformers, CI/CD
- Trained and deployed ML audience recommendation models | ZenML, MLOps, MLFlow
Senior Data Scientist — Technical Leader
FORVIA
- Led a cross-functional team of 6-8 (data scientists, software engineers, front-end developers) delivering ML solutions for manufacturing plants at world's 4th largest automotive supplier
- Built predictive maintenance models: process parameter optimization, time series forecasting, anomaly detection
- Applied LLMs for advanced text analysis on manufacturing documentation and reports
- Recognized internally as Machine Learning Expert
Data Scientist — PhD Researcher (CIFRE)
Orange
- Trained and deployed robust unsupervised anomaly detection models for real-time network traffic monitoring | PyTorch, PySpark, Docker
- Published 4 papers at top ML conferences: ECML-PKDD, IEEE COINS, AINA, SINConf
- Filed 2 international patents (INPI) for IoT anomaly detection and network traffic fingerprinting
- PhD thesis (CIFRE): "Robust Unsupervised Anomaly Detection"
Stack
AI / GenAI
LangGraphGoogle ADKLangChainRAGLangfuseMCP
vLLM
LiteLLMLoRA/QLoRAML / Deep Learning
PyTorchHuggingFace Transformers
scikit-learn
TensorFlowPEFTEngineering
PythonFastAPI
DockerKubernetes
KafkaSQLCI/CDCloud & MLOps
GCP / Vertex AIAWSPalantir
ZenMLMLFlow
Spark / PySparkLanguages
Projects
RagMaker — Agentic RAG Platform
Co-developed an Agentic B2B SaaS platform enabling users to query multi-format documents and connect external tools via conversational AI. Hybrid Search (semantic + keyword), Text-to-SQL, MCP integration, LangGraph agentic workflows.
LangGraphRAG
PostgreSQL pgvectorFastAPIMCPGCPDockermulti-agent-llm-stack
Coming SoonEnd-to-end production ML stack: multi-agent RAG with LangGraph, LLM fine-tuning (LoRA/QLoRA), vLLM serving, Docker, Kubernetes, Prometheus monitoring. Full stack, open source.
LangGraphRAGvLLMLoRAKubernetes
PrometheusPublications & Patents
RESIST: Robust Transformer for Unsupervised Time Series Anomaly Detection
2022N. Najari, S. Berlemont, G. Lefebvre, S. Duffner, C. Garcia
Robust Variational Autoencoders and Normalizing Flows for Unsupervised Network Anomaly Detection
2022N. Najari, S. Berlemont, G. Lefebvre, S. Duffner, C. Garcia
RADON: Robust Autoencoder for Unsupervised Anomaly Detection
2021N. Najari, S. Berlemont, G. Lefebvre, S. Duffner, C. Garcia
Network Traffic Modeling For IoT-device Re-identification
2020N. Najari, S. Berlemont, G. Lefebvre, S. Duffner, C. Garcia
International Patents
Contextual Anomaly Detection For The Maintenance Of IoT Devices
N. Najari, S. Berlemont, G. Lefebvre
Filed: June 2022, Orange Meylan
Assistance For The Identification Of Malfunctioning Devices Using Traffic Metadata
S. Berlemont, N. Najari, G. Lefebvre
Filed: November 2020, Orange Meylan
Contact
Interested in working together? Let's talk.