MBMohammed Benghanem.
Open to AI Engineer roles · On-site, hybrid or remote

I build intelligent products from model to interface.

Hi, I'm Mohammed Benghanem, a generative AI engineer who ships: RAG pipelines, QLoRA fine-tuning and agentic workflows, wrapped in secure APIs and polished React / Next.js interfaces.

View my workDownload CV
Mohammed Benghanem
AI Engineer & Full-Stack Developer
RAG · ClauseLens
95%
verified citations
False positives
82% → 0.1%
QLoRA fine-tuning
Multi-agent AI
LangGraph
Human-in-the-loop
Full-stack
Next.js
FastAPI · NestJS · Docker

Technologies I've shipped with

PyTorchHugging FaceClaude APILangGraphTensorFlowscikit-learnOllamaFastAPINext.jsReactPostgreSQLDockerAWS
About me

Generative AI, built for production

I'm an artificial intelligence engineer specialized in generative AI and web development, from the interface to the application services.

I design RAG systems, fine-tune large language models (LLMs) with QLoRA and orchestrate AI agents. I'm fluent in Python, FastAPI and React / Next.js to integrate predictive models and deploy secure applications.

  • RAG with hybrid search on pgvector
  • LLM fine-tuning with QLoRA (PEFT / TRL)
  • Multi-agent workflows & tool use
  • Secure APIs & Docker deployments
Get in touch
01

Generative AI

RAG with hybrid search, LLM fine-tuning (QLoRA, PEFT/TRL), multi-agent workflows with LangGraph and tool-using agents on the Claude API.

02

Full-stack engineering

FastAPI, NestJS and Node/Express backends, PostgreSQL / pgvector / PostGIS, and React / Next.js / TypeScript front-ends.

03

Production mindset

Leak-free evaluation, secured APIs with rate limiting, Docker deployments and real-time dashboards that people actually use.

Experience

Where I've been shipping

Three roles delivering AI and full-stack products for real businesses.

  1. Taxi TremplinCurrent role

    Freelance Web & AI Developer

    Aug 2026 — Present

    • Designed the full backend in NestJS, PostgreSQL / PostGIS and Redis: geolocated driver dispatch, fare calculation, SMS authentication and notifications. Deployed with Docker.
    • Built the admin dashboard in React / TypeScript to manage drivers, customers, rides, fares and payments.
    • Developed a smart booking assistant on the Claude API (Anthropic) that lets customers order a taxi in natural language through AI agents and tool use.
    NestJSPostGISRedisReactTypeScriptClaude APIDocker
  2. Yosobox

    AI & Web Development Intern

    Mar — Sep 2026

    • Developed two moisture-rate (LoD) prediction models: a physics-based model and an ANN model.
    • Integrated both models into a real-time dashboard: FastAPI microservice, secured Node/Express API and Supabase storage.
    • Created a secured API route (admin key, rate limiting) to trigger ANN predictions.
    • Fixed integration bugs and exposed the physics model's drying curve.
    PythonANNFastAPINode.jsExpressSupabase
  3. Lotus Capital

    AI & Web Development Intern

    Jan — Mar 2026

    • Built a social-media competitive intelligence platform: multi-source collection (X, LinkedIn, RSS, Reddit) and engagement analysis.
    • Designed multi-agent workflows with LangGraph to detect trends and generate content, with human-in-the-loop validation at every step.
    • Implemented embedding-based similarity search and duplicate detection (pgvector), comparing published posts against competitors'.
    LangGraphMulti-agentpgvectorEmbeddingsPython
Portfolio & case studies

Real projects, measurable impact

A selection of AI systems I designed, trained and deployed.

RAG · LLM fine-tuning★ Featured project

ClauseLens

Contract clause analysis with RAG & QLoRA fine-tuning

A RAG application that reviews commercial contracts against a 12-clause legal checklist, returning structured answers backed by exact quotations verified server-side.

  • Hybrid semantic + full-text search on PostgreSQL / pgvector, selected after benchmarking on a validation set.
  • Leak-free evaluation on 510 contracts annotated by legal experts.
  • QLoRA fine-tuning of Qwen2.5-1.5B (Hugging Face PEFT / TRL) on a single 6 GB GPU.
PyTorchQLoRAQwen2.5FastAPIPostgreSQLpgvectorNext.js

Results on unseen contracts

False positives

82%→0.1%

Before
After

Verified citations

69%→95%

Before
After

12

legal clauses

510

contracts

1.5B

parameters

Personal & open-source projects

Recall@1 on an adversarial benchmark

Vector only35%
Vector + re-rank58%
Hybrid (BM25 + vector)54%
Hybrid + re-rank100%
RAG · Research assistantOpen source

Academic Paper Q&A

Ask questions about research PDFs and get streamed answers grounded in the source text, with citations and a confidence score. Three-stage retrieval (vector + BM25 fused with Reciprocal Rank Fusion, then cross-encoder re-ranking) and a pluggable LLM: local Ollama, OpenAI or Anthropic.

FastAPIChromaDBBM25Cross-encoderReactDocker

35% → 100%

Recall@1, vector-only vs. full pipeline

View code

Evaluation on 300 held-out tickets

JSON parse rate100%
Schema valid96.7%
Category accuracy87.9%
Priority accuracy84.8%
LLM fine-tuning · SupportOpen source

Multilingual SaaS Support Agent

An EN/FR support agent that classifies SaaS tickets into 12 categories, predicts priority and sentiment, flags missing information and drafts a reply, all as strict, schema-validated JSON with automatic repair. Qwen2.5-3B fine-tuned with QLoRA on a 6 GB RTX 3060.

Qwen2.5-3BQLoRAPEFTFastAPIReactJSON schema

45% → 85%

Priority accuracy after rebalancing the dataset

View code

Model selection · 5-fold CV F1-macro

LinearSVC0.929
Logistic Regression0.919
Multinomial NB0.891
NLP · Data scienceOpen source

Job Market Insights

End-to-end study of the Moroccan job market: offers scraped from job boards like ReKrute, cleaned and explored by region and skill, then a classifier that predicts a job's category from its description and recommends matching university programs, served through FastAPI and React.

Web scrapingscikit-learnTF-IDFNLTKFastAPIReact

94%

Test accuracy across 9 job categories

View code
cloudsentinel · dashboard
CloudSentinel incident dashboard
Cybersecurity · Hybrid AIMaster's thesis

CloudSentinel

My master's final-year project: an AI intrusion detection system for cloud networks, trained on CSE-CIC-IDS2018 (16.2M flows, 77 features). A Random Forest stops known attacks, an autoencoder catches the zero-day anomalies it misses, and a local LLM (Llama 3.2 via Ollama) drafts grounded remediation reports for SOC analysts. Live traffic is captured with CICFlowMeter and streamed into a FastAPI + React dashboard.

Random ForestAutoencoderDQNOllamaFastAPIReactDocker

1.6% → 77%

Unseen attack families caught: Random Forest alone vs. autoencoder layer

From my professional experience

AI agent · MobilityFreelance

AI Taxi Booking Assistant

Taxi Tremplin

Natural-language ride booking powered by Claude agents and tool use, plugged into a geolocated dispatch backend with its own admin dashboard.

Claude APITool useNestJSPostGIS

3

Deliverables: API, admin, AI assistant

Multi-agent · Market intelInternship

Competitive Intelligence Agents

Lotus Capital

LangGraph multi-agent pipeline that tracks competitors, detects trends and drafts content with human review at every step.

LangGraphpgvectorHITL

4

Sources: X, LinkedIn, RSS, Reddit

Predictive ML · IndustryInternship

Moisture Prediction Dashboard

Yosobox

Physics-based and neural models for LoD prediction, served through a FastAPI microservice into a real-time dashboard.

ANNFastAPISupabase

2

Models: physics-based + ANN

Fine-tuning

95%

Verified citations on unseen contracts

Evaluation

510

Lawyer-annotated contracts, leak-free

Efficiency

6 GB

Single GPU used to fine-tune a 1.5B LLM

Experience

3

Industry roles in AI & web development

Hiring an AI engineer who takes models all the way to production?

Get in touch
Skills

My technical toolkit

From model training to production deployment.

Machine Learning & NLP

13
PyTorchTensorFlowscikit-learnPandasNumPyLLMsRAGFine-tuningQLoRALangGraphHugging FaceNLTKPrompt engineering

Backend & Databases

12
PythonJavaScriptNode.jsExpress.jsFastAPINestJSPostgreSQLpgvectorMongoDBChromaDBSupabaseRedis

Frontend

04
ReactTypeScriptNext.jsTailwind CSS

DevOps & Tools

04
DockerGitBashAWS
Background

Education & credentials

01

2024 — 2026

Master's in Big Data, Artificial Intelligence & Advanced Applications

Ibn Tofail University, Morocco

02

2021 — 2024

Professional Bachelor's in Mathematical Science & Computer Science

Ibn Tofail University, Morocco

Certifications

  • AWS Cloud Technical EssentialsAWS
  • Machine Learning with PythonIBM
  • Tools for Data ScienceIBM

Languages

  • ArabicNative
  • EnglishFluent (C1)
  • FrenchProfessional
Contact

Hiring an AI engineer? Let's talk

I'm available immediately for a full-time AI Engineer or Full-Stack Developer role — on-site, hybrid or remote, and open to relocation. The fastest way to reach me is by email, and my CV is available in English and French.

Emailmohammed.benghanem@uit.ac.ma
Phone+212 7 66 92 91 94
LocationTiflet, Morocco

Available for interviews

Send me the role or job description and I'll get back to you quickly with my availability.

Contact me by emailDownload CVAlso available in French