Open to data roles · Data Analyst / Scientist

Make data
actually useful.

I'm Rémy Grabet — I collect, analyse and visualise data, then build the full tools around it. Dashboards, ML models and automations that don't just sit in a notebook: the data work your team actually uses.

View my work Get in touch
Data science graduate 6 systems shipped Available now
🎓 Mines Saint-Étienne · Institut Polytechnique de Paris
▲
+126%
Portfolio tracked
Orbit — a data dashboard I built
7k
7,399
Facilities analysed
The stack I build with
PythonSQLPostgreSQLSupabasePandasscikit-learnPower BIFastAPIREST APIsGeoPandasPlotlyJupyterGitLinuxDocker-free PythonSQLPostgreSQLSupabasePandasscikit-learnPower BIFastAPIREST APIsGeoPandasPlotlyJupyterGitLinuxDocker-free
How I work

From raw data
to decisions.

Most data never makes it out of a spreadsheet. I take it the whole way — from scattered sources to a clean model, a clear dashboard, and a decision someone can act on. End to end, tested, and honest about what the numbers do and don't say.

Collect & clean

Scraping, APIs, entity resolution & fuzzy matching into a single clean source.

Analyse & model

Statistics, machine learning and rigorous out-of-sample validation.

Visualise

Interactive dashboards & maps that make the insight obvious.

Automate & deploy

Ship it as a running service — cloud Linux, scheduled, monitored.

Experience

Where I've done this for real

Data Science Consultant · Internship

BDO · Data & Analytics consulting
Mar – Sep 2025 · 6 months · France

End-of-studies internship: I built an end-to-end data platform mapping the French nursing-home (EHPAD) market for investors — from scattered public sources to a clean facility-level database, ML models and interactive decision-support dashboards.

7,399
Facilities mapped
86
Variables
×12
Faster pipeline
>94%
Entity match
  • Full ETL pipeline & multithreaded scraper (~4 h → ~20 min).
  • Fuzzy matching across 5+ public sources (>94% auto-match).
  • K-means segmentation, Random Forest / Gradient Boosting, Isolation Forest.
  • Interactive maps & a financial-prediction tool for the consulting team.
Pythonscikit-learnGeoPandasPandasWeb scrapingData viz
Toolbox

What I can do for you

Six things I do well — all drawn from projects I actually built, tested and deployed.

Data analysis & visualization

Turning messy data into clean dashboards, maps and reports that surface the story — the part decision-makers actually read.

PandasPower BIPlotlyDashboardsExcelData viz

Machine learning

Clustering, regression & anomaly detection — validated, not overfit.

scikit-learnK-meansRandom ForestIsolation Forest

Data pipelines / ETL

Scraping, cleaning & joining 5+ sources into one reliable dataset.

Web scrapingFuzzy matchingSQLPostgreSQLSupabaseSQLiteGeoPandas

Full-stack & apps

From FastAPI backends to interactive front-ends & desktop tools.

PythonFastAPIJavaScriptQt

Cloud, DevOps & rigor

Deploying to Linux cloud servers with systemd, monitoring and secrets kept safe — plus the scientific method that keeps results honest: hypotheses tested, biases hunted, ideas invalidated with numbers.

LinuxOracle CloudsystemdSSHGitOut-of-sample validation
Featured work

Projects I designed & built

A few personal projects taken end to end — from first idea to a working, deployed system.

🛰️ Flagship — a hub of data apps

Orbit

A personal data hub: trading, budget, planning & market news

A FastAPI launcher serving several HTML/CSS/JS dashboards in one hub — each turning raw data into charts, KPIs and projections. Plus a crypto-news watch summarised by a local LLM (Ollama), all running on my own machine.

6 apps
In one hub
Local LLM
On-device AI
9k→1.75k
Lines refactored
PythonFastAPIData vizOllama · LLMSQLite
View full details → ✦ Live demo ↗
Orbit — wealth projection Orbit — trading analytics Orbit — budget overview Orbit — capital tracker
🔑 Applied cryptography

KeyMint

Strong secrets, generated offline, never stored

A password & passphrase generator that runs entirely offline — on the phone (installable PWA) and on Windows (.exe). It shows the real entropy in bits rather than an arbitrary gauge, using unbiased rejection sampling proven with a chi-square test, and a French wordlist rebuilt from scratch across four linguistic signals.

99 bits
16-char entropy
χ² 500k
Proven bias-free
0 stored
Nothing persisted
JavaScriptWeb Crypto APIPWAPythonApplied cryptography
View full details → ✦ Live demo ↗
KeyMint — password mode with entropy in bits KeyMint — passphrase mode, rebuilt French wordlist
🛡️ Security — malware triage

VM Antivirus

Should I run this download? A verdict in three layers — without executing it

A desktop tool that decides whether an untrusted executable is safe, fusing static analysis, threat-intelligence reputation and behavioural sandbox detonation into one plain-language verdict. Suspicious files are detonated inside a disposable, isolated Windows VM to watch what they actually do — persistence, credential theft, network exfiltration — then the VM self-destructs.

3 layers
Static · reputation · sandbox
72+ AV
via VirusTotal
disposable VM
Behavioural detonation
PythonPE analysisVirusTotal APIWindows SandboxPowerShell
View full details → ✦ Live simulation ↗
VM Antivirus — malware verdict with threat intelligence VM Antivirus — sandbox detonation behaviour VM Antivirus — API quota and cost preview VM Antivirus — reputation API keys, kept local
More projects

The full toolbox

Other things I've built — from cloud bots to data-viz dashboards.

📊
Quantitative research

Polybot-Tracker

End-to-end research on prediction-market strategies: a full data pipeline & backtesting suite, reverse-engineering a pro bot from ~36k trades, validated out-of-sample.

5.5M tradesView details →
🎲
Cryptanalysis · Statistics

RNG Lab

A bench of 8 random generators, 8 statistical tests written from scratch, and an attack that predicts 100 % of Mersenne Twister's output.

✦ Live demo ↗
5 000/5 000 predictedView details →
📱
PWA + Windows · Sync

TDL — My tasks

A to-do list on phone & PC: one OKX design across a PWA and a native Windows app, offline-first Supabase sync, synced pinning.

✦ Live demo ↗
Supabase syncView details →
🤖
Deployed

Polycheat

On-chain copy-trading bot on a 24/7 cloud server — real-time detection over WebSocket, safe-by-design, systemd on Oracle Cloud.

~1.3s latencyView details →
⚖️
Research

Arbitrage & Fair-Value Lab

Latency-aware arbitrage & a Black-Scholes fair-value model — showing the apparent edge was an artefact, not alpha.

44h live dataView details →
🎯
Engine

Strategy Engine

Modular backtesting framework — market structure, FVGs, volume profile, metrics & validation, all unit-tested with pytest.

pytest testedView details →
🌾
IoT · Power BI

IoT — Rice Fields, Japan

Data visualisation from sensors implanted in rice fields in Kumamoto, Japan, with Power BI and Python.

Internship · JapanView details →
📈
Power BI

Investment Dashboard

A Power BI dashboard to monitor and analyse investments across markets — clean KPIs and breakdowns.

PersonalView details →
🌍
ELK Stack

Global Product Trends

A dashboard on the ELK Stack showing objectives, data distribution by country and key sales indicators.

Elasticsearch · KibanaView details →

Let's turn your data
into decisions.

Open to Data Analyst, Data Scientist and Data Consultant roles. The fastest way to reach me is email.

r.grabet@gmail.com LinkedIn GitHub
Legal notice

Legal notice

Last updated August 2026

Publisher

Rémy Grabet — France

Contact: r.grabet@gmail.com

Hosting

GitHub Pages — GitHub, Inc.

88 Colin P. Kelly Jr. Street, San Francisco, CA 94107, USA

Intellectual property

The content of this site (text, screenshots, source code) is © Rémy Grabet, unless otherwise stated. Third-party names and logos (technologies, projects) belong to their respective owners.

Personal data & cookies

This is a static showcase site: no account, no audience measurement, no tracking cookies. The only browser storage used is your theme preference (light / dark), kept locally on your device and never transmitted.