Mourad AbahaniAI Eng.
AvailableCV

AI Engineer — UM6P, Ben Guerir

Mourad Abahani, AI Engineer

I build conversational agents and LLM workflows that turn messy human intent into reliable, auditable action — shipped in production at UM6P, grounded in systems work from the 42 Network.

Start a conversation See the work CV

01About

I build AI systems that take messy human intent — a purchase request, a half-formed question — and turn it into reliable, auditable action. Agents that know when to ask a human. Evaluators that catch the model when it is wrong. Infrastructure that does not fall over.

I'm an AI Engineer designing and deploying conversational agents, LLM-powered workflows, and reinforcement learning systems in production.

My focus is procurement automationand enterprise AI integration, with multi-agent architectures at the core. The foundation is the 42 Network (1337 Coding School): intensive, peer-reviewed work in C, C++, Unix, and networking that taught me to build things that don't break.

Based
Ben Guerir, Morocco
Role
AI Engineer @ UM6P
Foundation
1337 · 42 Network
Languages
French · English

Or ask the machine — type help

02Experience

Where the work happened

  1. Mar 2025 — presentBen Guerir · MA

    AI Engineer

    Mohammed VI Polytechnic University (UM6P)

    • Engineered AIDA purchase agent — a production conversational AI procurement platform built on LangGraph with a Next.js frontend, integrating with enterprise RPA systems for real-time purchase requisition creation, routing, and validation.
    • Implemented Human-in-the-Loop (HITL) interrupt workflows, semantic catalog search, and automated PO form generation, reducing manual input and procurement errors across the department.
    • Advised senior operations leadership on AI adoption strategy and internal tooling automation — translating complex AI concepts for non-technical stakeholders through workshops and interactive demos.
    • Designed and shipped a reinforcement learning game for the DEO Open Day, managing the full pipeline from environment design and model training to live platform integration.
  2. Mar 2023 · 1 mo1337 · MA

    Volunteer — Think AI Hackathon

    1337AI

    • Organized and supported the Think AI Hackathon at 1337, facilitating communication between technical mentors and participating teams while gaining hands-on exposure to diverse AI applications through team projects.
  3. Aug 2024UM6P · MA

    Staff Volunteer — Xbox Game Camp Africa

    Microsoft

    • Supported event logistics and engaged with the developer community at Xbox Game Camp Africa 2024 at UM6P. Gained insights, expanded my network, and contributed to a successful collaborative environment.

03Selected work

Agents that ship, systems that hold

LIVE · PRODUCTION2025 — now

AIDA — AI Procurement Agent

A conversational AI platform that automates UM6P procurement workflows, from purchase request creation to parcours routing. A LangGraph agent pipeline classifies each article, searches framework contracts and supplier conventions via Meilisearch, and routes items through the correct procurement pathway (contrat cadre, convention, consultation, gré-à-gré, achat en ligne). Human-in-the-loop interrupts let users confirm matches before submission. An independent evaluator detects and auto-corrects agent hallucinations.

  • LangGraph
  • FastAPI
  • Next.js 15
  • PostgreSQL
  • Prisma
  • Redis
  • NextAuth.js
  • Vercel AI SDK
  • Docker Compose
  • NGINX
  • Loki
Read the build notes

Request lifecycle

  1. 01

    Intake

    A requester describes what they need in plain language through the Next.js + Vercel AI SDK chat.

  2. 02

    Classify

    A LangGraph agent splits the request into articles and classifies each one.

  3. 03

    Retrieve

    Framework contracts and supplier conventions are searched through Meilisearch.

  4. 04

    Route

    Each item is assigned its procurement parcours — contrat cadre, convention, consultation, gré-à-gré, or achat en ligne.

  5. 05

    Confirm

    A human-in-the-loop interrupt lets the user confirm matches before anything is submitted.

  6. 06

    Verify

    An independent evaluator detects hallucinated matches and auto-corrects them before the requisition is created.

Index — 6 more builds

  1. 002

    RL Game — DEO Open Day

    Designed and deployed an interactive reinforcement learning game for UM6P's Digital Ecosystem Office Open Day. Managed the complete pipeline — environment design, model training, and live platform integration for a public-facing audience demo.

    • Python
    • PyTorch
    • Reinforcement Learning
    2025
  2. 003

    Semantic Product Catalog Search

    Built a pgvector-based semantic search system for university procurement catalogs covering lab consumables and chemical products in French. Migrated from a memory-intensive JSON embedding approach to a PostgreSQL + pgvector architecture with a Next.js frontend — significantly improving search reliability and scalability.

    • Next.js
    • TypeScript
    • PostgreSQL
    • pgvector
    • LLM Embeddings
    2024
  3. 004

    ft_transcendence

    Front-end for a real-time multiplayer Pong website. Responsive SPA with user authentication, live chat, and tournament systems. Cross-browser compatibility and web security best practices.

    • Vanilla JS
    • HTML5 / CSS3
    • Bootstrap
    • WebSocket
    • REST
    42 Network
  4. 005

    Inception

    Docker-based infrastructure project. Created and managed multi-container applications using Docker Compose, implemented secure inter-service networking, and configured web servers, databases, and application stacks in isolated environments.

    • Docker
    • Docker Compose
    • NGINX
    • WordPress
    • MariaDB
    42 Network
  5. 006

    WebServer

    An HTTP/1.1 web server built from scratch in C++98 as part of the 42 curriculum. Supports GET, HEAD, POST, PUT, DELETE, static file serving, CGI for dynamic content, and concurrent clients via epoll().

    • C++
    • Sockets
    • epoll
    42 Network
  6. 007

    Minishell

    A custom Unix shell in C with process management, pipelines, I/O redirection, environment variable expansion, and robust memory management adhering to strict 42 coding standards.

    • C
    • POSIX
    • GNU Readline
    • Makefile
    42 Network

04Stack

Tools I reach for

04.1AI & Machine Learning

  • LangGraph
  • LangChain
  • RAG Pipelines
  • Multi-Agent Systems
  • HITL Workflows
  • Prompt Engineering
  • LLM APIs
  • Reinforcement Learning
  • PyTorch

04.2Backend & Data

  • PostgreSQL
  • Prisma
  • pgvector
  • FastAPI
  • Redis
  • RESTful APIs

04.3Frontend

  • Next.js
  • React
  • Tailwind CSS
  • shadcn/ui
  • Vanilla JS
  • HTML5
  • CSS3

04.4Programming Languages

  • Python
  • TypeScript
  • JavaScript
  • C++
  • C

04.5DevOps & Infrastructure

  • Git
  • Docker
  • Docker Compose
  • NGINX
  • Linux

04.6Collaboration

  • Cross-functional teamwork
  • Stakeholder comms
  • Workshops & demos
  • Peer review

05Education

Trained by peers, proven by code

1337 Coding School · 42 Network

Digital Technology Architecture · Software Engineering · 2022 — present

An intensive, project-based curriculum in systems programming, algorithms, and software architecture — C, C++, Unix, networking, and collaborative engineering through peer review.

06Contact

Let's build something that holds up in production.

Open to interesting problems in AI engineering, multi-agent systems, and production LLM workflows. Reach out — I read everything.

moradabahni@gmail.com