• Chief Engineer, E-halal
  • Master's, Université de Lille
  • Available for work-study

Fleury Niyokwizera · Data Engineer

4.5M+ products synced in 55 min instead of 9 h.

I build data pipelines, AI agents and open source platforms.

View my projects Download my CV

Live

  • halalopenfoodfacts.org
  • halalopenbeautyfacts.org
    Portrait of Fleury Niyokwizera
    Halal Assistant demo

    ⚠ AI assistant, it can make mistakes. Check the product label.

    I want halal milk, what do you suggest?

    I found 5 products for “halal milk”. Check the record before buying.

    • Goldium crémeux: compatible, halal label declared by contributors, not verified. View record ↗
    • Code 6111259092525: ambiguous. Sensitive ingredient: rennet. View record ↗

    Try:

    Demo: answers rebuilt from real product records. Links open the real records.

    • 4.5M+

      products synced

    • 350

      registered contributors

    • 12,000

      visitors per month

    • 99.5%

      availability

    About

    I build the data chain end to end.

    I am a Data Engineer. Ingestion, modeling, deployment, reporting: I own the whole chain, and I measure it.

    As E-halal's volunteer Chief Engineer (Oct. 2025 – June 2026), I designed Halal Open Food Facts and Halal Open Beauty Facts: architecture, import of 4.5M+ products, self-hosted infrastructure. I then integrated Halal Assistant, an AI agent that queries the database in plain language.

    I am a first-year Master's student in Data Modeling at Université de Lille, after a Master's in Applied Statistics and Business Intelligence and a Bachelor's in Actuarial Science at ISTA, University of Burundi.

    Fleury Niyokwizera working at a desk in front of several screens of code

    Key figures

    Halal Open Food Facts and Halal Open Beauty Facts platforms, in production.

    • 4.5M+

      products ingested and synchronised automatically

    • 12,000

      visitors per month

    • 25,000

      Celery tasks processed per day

    • 99.5%

      availability

    • < 300 ms

      median API response time

    • 9 h → 55 min

      sync time, a 90% reduction

    Projects

    Data in production, not in a demo.

    Featured project · AI agent

    Halal Assistant: ask instead of search.

    An AI agent built into Halal Open Food Facts and Halal Open Beauty Facts. It queries the product database through RAG and answers in plain language, with a nuanced status and a link to the source record.

    • < 3 s

      response time

    • 85%

      accuracy on 200 test questions

    • 4.5M+

      products searchable in plain language

    Halal Assistant halalopenfoodfacts.org

    ⚠ AI assistant, it can make mistakes. Check the product label.

    I want halal milk, what do you suggest?

    I found 5 products for “halal milk”. Product names don't always make it clear that it is milk, so check the record before buying.

    • Goldium crémeux: compatible based on the listed ingredients, but the halal label was declared by contributors and is not verified. View record →
    • Maître Fromager: compatible, halal label declared by contributors, not verified. View record →
    • Code 6111259092525: ambiguous. An ingredient of uncertain origin or conflicting labels. View record →
    Reconstruction of a real exchange on halalopenfoodfacts.org, translated from French (question rephrased, answer shortened).
    1. UserQuestion in plain language
    2. Halal Assistant widgetOn both sites
    3. RAG retrievalpgvector · PostgreSQL
    4. LLMClaude via OpenRouter
    5. AnswerNuanced status + source record
    Simplified architecture: answers are grounded in the records retrieved from the database.

    What the agent does

    • Answers questions about a product.
    • Assesses halal compliance from the ingredients.
    • Helps users choose halal products that fit their request.
    • Helps complete a product record.
    • Guides users through the sites.

    Safeguards

    • Permanent warning: the AI can be wrong, check the label.
    • Nuanced status: compatible, ambiguous, label verified or only declared.
    • Every product cited links to its source record.

    Stack

    • Python
    • RAG
    • pgvector
    • PostgreSQL
    • Claude
    • OpenRouter

    halalopenfoodfacts.org halalopenbeautyfacts.org See the real screenshot ↗

    The database in numbers

    Taken from both sites in October 2026. Halal + non-compliant + doubtful + unknown = total.

    Halal Open Food Facts 4,571,162 products

    Halal
    1,097,660
    Non-compliant
    103,048
    Doubtful
    111,759
    Unknown
    3,258,695

    Halal Open Beauty Facts 67,759 products

    Halal
    14,860
    Non-compliant
    3,683
    Doubtful
    612
    Unknown
    48,604

    Other projects

    • Data Engineering · E-halal

      Halal Open Food Facts & Beauty Facts platforms

      Two open source data platforms, from ingestion to infrastructure. 4.5M+ products, 12,000 visitors a month.

      • Django REST
      • PostgreSQL
      • Redis / Celery
      • Docker
      • Nginx
      • OVHcloud
      Details
      • Streamed import of the Open Food Facts and Open Beauty Facts CSV dumps into PostgreSQL, then hourly delta sync (Celery Beat).
      • Sync time cut from 9 h to 55 min (−90%); 25,000 Celery tasks processed per day.
      • Secured infrastructure on an OVHcloud VPS (HTTPS, backups, restricted access): 99.5% availability, median API response < 300 ms.
      • Open source contribution process (CI, documentation, governance): 350 registered contributors, 40 pull requests merged.

      GitHub · Food ↗ GitHub · Beauty ↗

    • Big Data · ISTA

      NYC Yellow Taxi Big Data pipeline

      40 million rides processed with PySpark, 6 times faster than with pandas.

      • PySpark
      • Parquet
      • scikit-learn
      • Streamlit
      Details
      • PySpark pipeline on Parquet data: 40 million rides, processing time divided by 6 compared with pandas.
      • Four machine learning models compared; best model R² = 0.87.
      • Interactive Streamlit dashboard to report the indicators.

    • Machine Learning

      Insurance claims prediction

      Modular ML pipeline: 5 models compared, best AUC = 0.81.

      • Python
      • pandas
      • scikit-learn
      • Joblib
      Details
      • Hyperparameter search (RandomizedSearchCV), 5-fold cross-validation, configurable dataset mapping.
      • Model serialisation with Joblib, ready for production.

      GitHub ↗

    • Document AI

      Documents Assist

      Query your documents in plain language, with verifiable sources.

      • Python
      • Streamlit
      • PyMuPDF
      • TF-IDF
      • Optional LLM
      Details
      • Multi-format extraction (PDF, DOCX, XLSX, EPUB, HTML…) and page-level chunking.
      • Hybrid word/character TF-IDF search; citations built from metadata, never from generated text.
      • Runs locally without an API key; optional LLM through an OpenAI-compatible API.

      GitHub ↗

    • First-year Master's project · ISTA

      OpenISTA

      University management platform for ISTA, University of Burundi.

      • Flask
      • PostgreSQL
      • SQLAlchemy
      • Jinja2
      Details
      • Flask/PostgreSQL application: students, faculty, courses, library and internal messaging.

      GitHub ↗

    • Full-stack · Marketplace

      Maghreb Market

      Classified ads marketplace for North Africa.

      • Next.js
      • TypeScript
      • Django
      • PostgreSQL
      Details
      • Geographic search, guided listing creation, buyer/seller messaging, automatic moderation and back-office.

      GitHub ↗

    • Applied statistics

      ISOKO IRAMA / SESA study

      Statistical analysis of 489 respondents for an EU-funded project.

      • R
      • SPSS
      • KoboToolbox
      Details
      • Methodological split between organisations and individuals, and field data collection via KoboCollect.

    • Personal project

      Fleury Finance Tracker

      Personal finance tracker in BIF/EUR, in a single HTML page.

      • JavaScript
      • GitHub Gist API
      Details
      • Data synchronised through the GitHub Gist API.

      GitHub ↗

    Experience

    From statistics to data in production.

    Work experience

    1. Oct. 2025 – June 2026 · Lille

      Volunteer Chief Engineer, open source platforms

      E-halal · Halal Open Food Facts & Halal Open Beauty Facts

      • Designed and built two data platforms in production (Django REST, PostgreSQL, Docker, Redis/Celery): 4.5M+ products, 12,000 visitors/month, 350 registered contributors.
      • Migrated the infrastructure to an OVHcloud VPS and automated the import: sync time cut from 9 h to 55 min (−90%), 25,000 Celery tasks per day.
      • Secured the infrastructure and set up open source contribution: 40 pull requests merged, 99.5% availability, median API response < 300 ms.

    2. Feb. – June 2025 · Burundi

      Assistant consultant, project monitoring and evaluation

      UMWUGA N'AKAZI project (CREOPS, BBIN, WHH)

      • Collected and analysed field data with KoboToolbox: 1,200 households surveyed, 45 indicators tracked, 6 quarterly reports.
      • Automated data cleaning and consolidation: report preparation cut from 3 days to 4 hours.

    3. July – Oct. 2024 · Burundi

      Intern, Data Science applied to insurance

      SOCAR Vie

      • Prototyped AI in life insurance pricing: 50,000 contracts, gradient boosting, prediction error 12% lower than the baseline model.
      • Presented the results to technical management: recommendations adopted for the pilot phase.

    Education

    1. 2026 – 2027

      Master 1, Data Modeling

      Université de Lille

    2. 2025 – 2026

      Master 2, Applied Statistics and Business Intelligence

      ISTA, University of Burundi · thesis being finalised

    3. 2024 – 2025

      Master 1, Applied Statistics and Business Intelligence

      ISTA, University of Burundi

    4. 2021 – 2024

      Bachelor's in Actuarial Science, Risk Analysis

      ISTA, University of Burundi

    Skills

    Tools, by purpose.

    • Data Engineering

      • Python
      • pandas
      • PySpark
      • SQL
      • Bash
      • ETL/ELT pipelines
      • Database synchronisation
      • Big data and real-time processing
      • Data modeling
      • Redis / Celery

    • AI & LLM

      • LLM integration via API
      • Conversational agents
      • RAG
      • pgvector
      • Claude
      • OpenRouter

    • Databases

      • PostgreSQL
      • Oracle
      • NoSQL
      • Redis

    • Machine Learning

      • scikit-learn
      • Predictive models
      • Cross-validation
      • Gradient boosting

    • Infra & DevOps

      • Docker
      • Linux
      • OVHcloud
      • GCP
      • Nginx
      • Git / GitHub
      • GitHub Actions

    • BI & statistics

      • Power BI
      • Streamlit
      • R
      • Excel / VBA
      • Stata
      • SPSS
      • KoboToolbox

    • Web & API

      • Django REST
      • Flask
      • React
      • Next.js
      • TypeScript

    Contact

    Let's talk data.

    Looking for a Data Engineer, AI Engineer, Data Analyst, Data Scientist or BI work-study student or intern from November 2026? Get in touch.

    Download my CV