Skip to content
FR

Hello, I'm Mohamed

Engineering intelligent systems,from research to real software.

AI, Data & Software Engineer. I build applied AI systems, reliable data pipelines and full-stack software, with rigorous evaluation and production-minded engineering.

  • LLM
  • RAG
  • Pipelines
  • Real time
  • Next.js
Explore my work CV

Fig_01 — one field, three representations

Engineer pass
Mohamed AbiMA_2026

Mohamed Abi

AI · Data · Software

PAR · FR
Profiles@

Paris, France

Open to a permanent role in France

—:—Paris · local time

Flagship work

TotalEnergies OneTech
LLM extraction on GPU/HPC
CETIC
RiskTailor, INFORSID 2026 — first author
Maison Dlala
live real-time platform
Paris-Saclay
MSc DataScale
Mohamed Abi01 — 03

Three systems, three kinds of proof.

An industrial evaluation, a research publication, a live product. Each case study states what I personally did — and what isn't mine.

  1. TotalEnergies OneTech · DataLab, Pau

    The right value, for the right object: LLM extraction from industrial reports

    Two-pass extraction guided by document structure, on self-hosted LLMs and GPUs, with an evaluation that separates identifying the objects from the quality of their attributes.

    Evaluated prototype, candidate for integration

    Read the case study

    REPORT · 175K TOKENS§1 WELL A§3 WELLBORE A-1§7 OFFSET WELLWELL AWELLBORE A-1OFFSET — EXCLUDEDCONCEPTUAL — FICTIONAL VALUES
    Concept_01 · illustration
  2. Maison Dlala live sale room, buyer view
    Live

    Independent product · real time, multi-instance

    Maison Dlala: live auctions where the server is always right

    A four-language auction marketplace, Arabic included: idempotent bids inside one atomic write, a server-owned clock, live rooms spread across several instances.

    Private showroom, v2.6 — fictional houses, lots and bidders, simulated payments

    Live — maisondlala.com

    Read the case study

  3. CETIC · NLP, GenAI and cybersecurity

    RiskTailor: one risk report, one rendition per reader

    A RAG pipeline behind an API that tailors a cyber-risk report summary to the architect, the manager or the compliance lead — published at INFORSID 2026, as first author.

    Published research — INFORSID 2026

    Read the case study

    Reader 01ArchitectVulnerabilities, exposed componentsSame report
    Reader 02ManagerImpacts and prioritiesSame report
    Reader 03ComplianceISO 27005 traceabilitySame report
All work and the archive

02Work experience

Industry, research, product.

  1. 04 — 09 · 2026

    TotalEnergies OneTechHPC Data Scientist / Data Engineer (MSc internship)

    Metadata extraction with self-hosted LLMs on GPUs, with an evaluation that separates identifying objects from the quality of their attributes.

    Case study

  2. 01 — 03 · 2026

    CETICNLP & GenAI R&D, cybersecurity

    RAG pipeline and FastAPI service tailoring a risk report to its reader; first author of RiskTailor (INFORSID 2026).

    Case study

  3. 2025 — 2026

    Maison DlalaFounder — product and architecture

    Real-time auction platform in production: idempotent bids, multi-instance live rooms, four languages including Arabic.

    Case study

  4. 06 — 08 · 2024

    CERISTVideo processing & computer vision

    Video anomaly-detection pipeline: ingestion, features, CNNs and autoencoders, evaluation.

  5. 2023 — 2025

    WorldWeb SolutionsFounder, full-stack developer

    End-to-end web products for small businesses: scoping with non-technical clients, build, launch.

  6. 2022

    Sonatrach · APCSData & enterprise applications

    Oracle modelling and PL/SQL, an APEX application with row-level security; Odoo modules and SQL logistics reporting.

03Skills

Three disciplines, one proof per line.

No gauges, no percentages: every capability points to the work where it was used.

04How I work

Five rules, each learned on a real problem.

  1. 01

    Evaluate before claiming

    An average score hides the errors that cost. I separate what I measure — detection, attribution, latency — and keep the limits in the same document as the results.

    L1 / L2 evaluation, TotalEnergies

  2. 02

    Put the guarantee in the write

    When a rule matters — a bid never counted twice — it lives in the atomic operation, not in surrounding code that a crash can bypass.

    Idempotent bids, Maison Dlala

  3. 03

    The smallest safe design

    Between the cleanest architecture and the one that leaves the money path alone, I pick the second — and write down why.

    Live-sale adapter, Maison Dlala

  4. 04

    A generator, a checker

    An LLM proposes, a verifiable constraint decides: an ontology, a Pydantic schema, a deterministic validator.

    Neuro-symbolic graphs · TotalEnergies validators

  5. 05

    State the exact status

    Prototype, research, built but switched off, in production: these are different states, and blurring them costs trust.

    This whole site

05Research & education

To publish is to accept being checked.

Publications

Education

  1. 2025 — 2026

    MSc DataScale — big data in large distributed systems

    Université Paris-Saclay / UVSQ

  2. 2024 — 2025

    MSc year 1 — Big Data Management

    Université Paris-Saclay / UVSQ

  3. 2023 — 2024

    MSc Data Science

    Université d'Alger

  4. 2020 — 2023

    BSc Information systems & software engineering

    Université d'Alger

Certifications & courses

  • Snowflake — Associate Data Engineer2025
  • DataCamp — Data Engineer Associate2025
  • Dataiku — ML Practitioner · Core Designer2025
  • Databricks — Introduction to Databricks SQL2026
  • AWS — Data Engineering Foundations · Cloud Technical Essentials2025
Portrait of Mohamed Abi
Mohamed Abi · Paris

06About me

An engineer between research and software.

I started with information systems in Algiers: databases, PL/SQL, enterprise applications. Then pipelines and machine learning, then applied generative-AI research at Paris-Saclay, CETIC and TotalEnergies.

What keeps me interested are problems where a good answer isn't enough: it has to be verifiable, attributed to the right object, and still hold when everything happens at once. That's as true of an LLM reading a 175,000-token report as of a bid sent twice during a live sale.

I also explain data and AI to a community of over 100,000 followers. I speak Arabic, French and English.

Hiring

An AI, data or software role? Let's talk.

Looking for a permanent role in France. Reply within 48 hours.

contact@mohamedabi.com LinkedIn GitHub