NICOLAS DECOOPMAN

Data Scientist — Whympr

Remote · Apr 2026 – Present

Studies and prototypes on snow cover modeling and surface data fusion for mountain environmental products.

Modeling & data

• Temporal modeling to estimate and track snow conditions from heterogeneous sources.
• Fusion and preparation of geospatial datasets for training and evaluation pipelines.
• Field validation to calibrate models and monitor prediction quality.

Stack: Python · scikit-learn · Machine Learning · GIS

Data & Analytics Engineer — Whympr

Remote · Feb 2026 – Apr 2026 · 3 months

Implementation of a centralized analytical infrastructure. The primary goal of this project was to unify behavioral, transactional, and product data into an industrialized daily pipeline to provide teams with a comprehensive view of user activity.

Data engineering & architecture

• Development of a Data Warehouse based on DuckDB with a 4-layer ELT approach (Staging, Clean, Build, Marts) managed end-to-end via dbt.
• Implementation and maintenance of Python extraction pipelines targeting 4 main data sources: Google Analytics 4 (BigQuery), RevenueCat (via Amazon S3 and API), and the PostgreSQL production database.
• Implementation of an identity resolution logic aiming to reconcile non-logged session paths with application user IDs.

Modeling & analytics

• Creation of analysis models covering over 180 GA4 events, categorized by major usages (Navigation, Social, Monetization, etc.), to facilitate product KPI tracking.
• Modeling the subscription lifecycle (trial, conversion, renewal, cancellation) by algorithmically managing continuous periods (via an Islands & Gaps logic).
• Construction of One Big Tables (OBT) serving as a sustainable foundation for user segmentation and cohort studies.

BI & business adoption

• Deployment of an Apache Superset instance via Docker to allow Product and Marketing teams to explore data independently, with guidance on key reports (retention by cohort, Premium conversion funnels).
• Implementation of deployment on VM with a containerized stack, automation of service starts, and an operational runbook to secure production deployments and daily maintenance.

Stack: Python · dbt · DuckDB · PostgreSQL · SQL · Docker · Apache Superset

Full-Stack Data Engineer — naivo

Remote · Sept 2025 – Feb 2026 · 6 months

Design, development, and operation of a complete digital product. From several disparate sources to a single mountain decision-support platform. naivo.fr resolves the fragmentation of snow-weather safety data by centralizing official and collaborative streams on a high-performance map interface.

Product & value chain

• Mastery of the entire value chain: from data ingestion to the final user experience.

Data engineering & ETL

• Asynchronous ingestion (Python) of JSON/XML streams (Météo-France, BERA, Skitour, Camptocamp) with cleaning and spatialization pipelines.

GIS infrastructure & performance

• Serverless GIS Infrastructure: design of a pipeline (GDAL, Tippecanoe, PMTiles) processing digital elevation models at 25m resolution.
• High availability & frugality: distribution of vector tiles and rasters via Cloudflare Workers and R2 (PMTiles format), ensuring < 100ms latency on mobile.

Diffusion & recognition

• Social media automation: CI/CD pipelines (GitHub Actions) driving automated interface capture (multi-layer, multi-format) and its publication on Instagram and Facebook, daily illustrating the platform's functional richness and diverse use cases.
• Institutional recognition: project selected and honored by data.gouv.fr among the remarkable Open Data reuses of the year, validating the technical quality and product relevance.
• Media coverage: covered by regional (Le Dauphiné Libéré) and specialized press (La Belle Route) for its concrete impact on public mountain safety.

Stack: Python · SQLite · GDAL · Cloudflare Workers · Astro/React · MapLibre GL JS · GitHub Actions

Data Engineer & Scientist — CNRS / IGE / Météo-France

Grenoble, France · Mar 2025 – Sept 2025 · 7 months

Complete industrialization of the analysis of hourly extreme precipitation in France (1959–2022): from massive climate data processing to the production of automated scientific deliverables.

Big data engineering

• Initial design of a distributed ETL pipeline (Dask, Airflow, Linux HPC cluster) processing 10 billion data points (100 GB) across 88,000 modeled points and 14,000 observed points covering 560,640 hours. Optimized formats: NetCDF - Zarr - Parquet. Performance gain of 80% vs sequential execution.

Advanced statistical modeling

• Automatic selection engine among 7 spatio-temporal GEV models (stationary and non-stationary) via profiled log-likelihood, applied point-by-point for seasonal and monthly extremes characterization.

DevOps/MLOps & reproducibility

• Full CI/CD chain (GitHub Actions, Docker) with automated versioned dataset publication on HuggingFace Datasets, Streamlit application deployment on HuggingFace Spaces, interactive dashboards (Plotly, Leafmap), and dynamic scientific reports (Quarto). Configuration-as-code (YAML), monitoring, and total reproducibility guaranteed.

Stack: Python · Dask · Xarray · Polars · NumPy · SciPy · Numba · Zarr · Parquet · Airflow · Docker · Linux HPC · Streamlit

Doctor of Veterinary Medicine — Clinique vétérinaire des Dômes

Le Broc, France · Sept 2021 – Aug 2024 · 3 years

Mixed practice in canine and rural medicine, combining clinical care, herd monitoring, and a scientific approach to animal health data.

Responsibilities & achievements

• Emergency care and clinical prioritization (canine and rural), with standardized protocols and complete traceability.
• Coordination and monitoring of technical and scientific projects on livestock health, nutrition, and reproduction.
• Rigorous methodology: diagnostic reproducibility, protocol compliance, patient data management.
• Pedagogical communication and scientific popularization for owners, technicians, and agricultural partners.
• Operational collaboration with health authorities, laboratories, field actors, and international institutions (One Health logic).

Doctor of Veterinary Medicine — Clinique Ani-Médic

La Tardière, France · Sept 2019 – Aug 2021 · 2 years

Sanitary and technical monitoring of cattle, sheep, and goat farms with an integrated animal-environment-production approach.

Responsibilities & achievements

• Analysis and valuation of field data (reproduction, nutrition, pathologies, welfare).
• Sanitary and preventive plans based on measurable indicators (herd monitoring, fertility, milk quality).
• Interdisciplinary coordination between veterinarians, technicians, and farmers for complex issues.
• Training and support for farmers, synthetic technical reports for decision-making.

Data Scientist & Analyst — INRA

Le Rheu, France · Mar 2019 – July 2019 · 5 months

Design of operational lactosemia indicators for livestock consulting organizations, based on statistical and mechanistic modeling of lactose transfers between the udder and the blood.

Analytical rigor

• Management of quality control and analytical validation on 766 measurements from 10 experiments (279 cow profiles). Result: CV ≤ 3% (repeatability) and ≤ 7% (reproducibility), ensuring indicator reliability for production integration.

Advanced mechanistic modeling

• Development of a two-compartment udder-blood model quantifying complex physiological transfers (2.1 g/12h of milking; 130 g/24h; renal clearance $0.24 L \cdot min^-1$). Evidence that in case of inflammation, up to 22% of lactosemia variations are explained by animal profiles.

Decision-making impact

• Translation of experimental data into robust indicators directly integrable into decision-support tools for livestock consultants.
• Publication at the Académie Vétérinaire de France and ADSA-INRAE communication.

Stack: R · lme4 · lmerTest · emmeans · ggplot2

Veterinary Assistant — Clinique Ani-Médic

Moncoutant, France · Jan 2019 · 1 month

Bovine prophylaxis campaign.

Field mission

• Collection and entry of sanitary data in the field, standardization of records, quality control, monitoring of bovine screening indicators.
• Collaboration with veterinary teams and state services.

FeelingsAnalysis

Nov 2025

Multi-aspect AI system for automating customer feedback analysis (price, cuisine, service, atmosphere) on restaurant reviews in French. Generalizable to e-commerce, satisfaction surveys, and HR.

• Strategic benchmark: rigorous comparison between zero-shot LLM approach (Ollama) and fine-tuning a CamemBERT-Large transformer model (110M parameters) with 4 independent classification heads. Over 24 iterations of experiments to reach the optimal configuration.
• GPU optimization: reproducible training pipeline (PyTorch Lightning) integrating Mixed Precision (FP16, 2–3x acceleration), Gradient Checkpointing (−40% GPU memory), Gradient Accumulation (effective batch 128), and Discriminative Learning Rates. Result: 86% macro-accuracy (87.8% on service, 87.2% on cuisine).
• Production-ready: modular architecture, centralized configuration, and automatic logging of metrics, designed for industrial deployment via REST API.

Stack: Python · PyTorch Lightning · Hugging Face Transformers · Ollama · scikit-learn

SunCast

Nov 2025

High-Performance Computing (HPC) simulation engine for high-resolution sunrise and sunset times, integrating shadows cast by real topography (Copernicus DEM).

• High-Performance Computing: massively parallelized C++17 calculator (OpenMP, 96 cores per task) processing ~2 million pixels × 365 days per department. C++ → Python binary communication in streaming (zero intermediate files). Deployment on HPC cluster via Slurm Job Arrays.
• Use cases: actionable data for solar panel placement optimization in mountains, glaciological studies, precision agriculture, and urban planning.

Stack: C++17 · OpenMP · Python · NumPy · GDAL · Slurm HPC · Parquet

Nivéo

Oct 2025 – Nov 2025

100% autonomous and serverless data infrastructure architecture for real-time snow monitoring in France, using public Météo-France APIs (DPClim).

• Total automation & DevSecOps: ingestion and visualization pipeline running daily without human intervention. Enhanced security: AWS authentication via OIDC (zero long-lived keys), encrypted secrets, minimal IAM permissions.
• Infrastructure as Code (IaC): Terraform provisioning for AWS (Lambda, DynamoDB with 11-day TTL). Zero operational cost stack (permanent Free Tier) with automatic cleaning and GitHub Actions exports.
• Performant frontend: lightweight static site (Astro, MapLibre GL JS, Chart.js) automatically deployed to GitHub Pages with every data update.

Stack: Python · AWS (DynamoDB, Lambda) · Terraform · GitHub Actions · Astro · MapLibre GL JS · Chart.js

MissingDataLab

Oct 2024 – Feb 2025

Methodological comparative study of missing data imputation strategies (MCAR, MAR, MNAR) to guarantee predictive model robustness in real-world contexts.

• Rigorous benchmarking: systematic evaluation of 7 methods (mean, median, KNN, SoftImputer, PCA, ICE, MissForest) on simulated scenarios, demonstrating MissForest and ICE's superiority in maintaining statistical integrity.

Stack: Python · scikit-learn · SciPy · pandas · NumPy · R (simstudy)

BioResistanceAI

Jan 2025

Predictive AI for antibiotic resistance: predicting bacterial resistance to 5 antibiotics from multi-omic data (414 bacteria, 94,000+ features: 72,236 SNPs, 16,005 genes, 6,026 gene expressions).

• Predictive excellence: recall of 0.96 (Tobramycin via XGBoost) and stable performance across all antibiotics via ensemble model optimization (XGBoost, LightGBM) with systematic hyperparameter tuning (GridSearchCV, 5-fold cross-validation).
• Explainable AI & feature importance: identification of the most predictive information sources per antibiotic (dominant transcriptomics for the best models), to guide rapid diagnostic strategies in clinical contexts.

Stack: Python · scikit-learn · XGBoost · LightGBM · PyTorch · pandas · NumPy

SnowTrack

Dec 2024 – Feb 2025

Orchestration of automated validation pipelines for climate forecast model evaluation (S2M vs field observations from Météo-France).

• ETL automation & spatio-temporal analysis: automated flows for data ingestion, processing, and aggregation, generating detailed statistics (means, maxima, distributions, trends) and identifying zones and periods with significant discrepancies between models and observations.

Stack: Python · pandas · NumPy

DevOps — Portfolio {NCS}decoopman

2024 - Present

Automation of personal site publication, CV, and analytical reports (Quarto) within a reproducible chain.

CI/CD & publication

• Complete automation of site and analytical report publication.
• Every commit triggers site regeneration (Astro) and reports (Quarto), generation and upload via GitHub Actions.
• CI/CD ensuring consistency, traceability, and editorial time savings.

Stack: Astro · TypeScript · Quarto · GitHub Actions