Filter my work by industry, search it, and click straight into live demos.
Business + data-science background. I build at the seam where strategy meets a working product — where an analysis stops being a notebook and becomes a system someone actually runs. My work spans life sciences, cybersecurity, supply chain, energy, HR, finance, and sports — same craft, different domain.
- 🌍 Lebanon · UAE · GCC · Europe
- 💼 Strategy · Product · Data Analytics · Applied AI
- 🎯 Interested in: process intelligence, operational AI copilots, computational biology & fintech
- 🔗 Live portfolio: jadzoghaib.github.io
| Project | What it does | Stack |
|---|---|---|
| 🟢 Genetic Variant Triage · code | Thousands of genetic variants sit unclassified — not benign, not pathogenic, just uncertain. Reconciles AlphaMissense's predictions against ClinVar's clinical assertions on AlphaFold structure, then grades how much evidence actually backs each call. Rebuilds weekly against live ClinVar | AlphaFold AlphaMissense ClinVar DuckDB 3Dmol.js |
| ↳ Illustrated manual | The problem, the data sources, and how the triage actually works — written for a reader who has never seen a variant table | SVG Explainer |
| 🟡 Batch Investigation Console · code | Traces tablet quality back to raw-material attributes and compression conditions across 1,005 real production batches. Built around one finding: 22 of 44 candidate drivers reverse their correlation sign pooled vs. within a peer cohort — so every comparison is cohort-scoped | Python Streamlit pandas RCA |
| 🟢 EnzySelect · code | Prioritizes candidate enzymes for PET plastic degradation against operating conditions and a testing budget — transparent, decomposable scoring with real AlphaFold/UniProt structural anchors. Synthetic candidate set, and the app says so on every screen | Python Streamlit AlphaFold UniProt |
| Breast Cancer Detection | Cloud-deployed, containerized ML classifier — final cloud-computing project | Python AWS ML |
| Project | What it does | Stack |
|---|---|---|
| 🟢 ProcessLens · code | Celonis-style purchase-to-pay process intelligence on a 1.6M-event log — process discovery, throughput funnels, conformance checks, spend-weighted opportunity board | React TypeScript pandas Process Mining |
| Supply Chain Sentinel | Autonomous supplier-risk watch: GDELT + LLM analysis, scheduled sweeps, geospatial impact scoring, approval-gated client advisories | FastAPI n8n GDELT LLM |
| 🟡 GroceryAI · code | Full-stack nutrition and grocery app: meal planning, LLM shopping-list generation against real Mercadona supermarket data, a fridge-to-recipe engine, a body optimizer, a health feed, and a multi-agent basket debate — all one SPA over a FastAPI server | FastAPI LangGraph Groq LLM Python |
| Project | What it does | Stack |
|---|---|---|
| 🟡 Aegis — SOC Triage Copilot | Multi-source alert ingestion → threat-intel enrichment → incident correlation → dual-classifier severity triage → human-gated containment, every stage a visible n8n workflow | n8n FastAPI SQLite MITRE ATT&CK |
| ↳ Documentation hub | Four references: the interactive threat manual (14 techniques mapped to MITRE ATT&CK, each with a clickable architecture diagram), the six workflow specs, the ontology, and the incident lifecycle | SVG ATT&CK |
| Project | What it does | Stack |
|---|---|---|
| TalentFlow | n8n-orchestrated hiring funnel: LLM profile extraction, cold-start scoring, recruiter review gates that resume paused workflows, weekly calibration | Node 24 n8n LLM |
| 🟢 TeamMatch · code | Hybrid recommender staffing engineers to projects at portfolio scale (matrix factorization + content + personality) with a cohesion-aware optimizer | React Recommender Optimization |
| Project | What it does | Stack |
|---|---|---|
| OilShield | Strategic stress-testing console for oil & gas portfolios: build a company from typed assets, shape a 36-month Brent scenario, watch break-evens, liquidity headroom and shut-in economics recompute. Revenues float with price while fixed costs keep running — that asymmetry is the product | Python Streamlit Plotly |
| Project | What it does | Stack |
|---|---|---|
| 🔒 VaultTech — Forging Line | End-to-end cloud ML pipeline for an industrial forging line: medallion data architecture → XGBoost on SageMaker → served via ECS/Fargate → Streamlit ops dashboard | AWS SageMaker ECS/Fargate XGBoost Streamlit |
| Project | What it does | Stack |
|---|---|---|
| Stride | Athlete monetization and sponsorship platform: athletes own evidence-based marketability analytics, sponsors match campaign briefs with fully decomposable scoring, deal pipeline runs end to end | FastAPI React Supabase |
| 🟢 NIL Monetization Platform · code | Athlete name-image-likeness platform with AI-assisted sponsorship matching between athletes and brands | React TypeScript AI Matching |
| 🟢 Movie Recommender · code | MovieLens recommender with an EDA archive, a drift-simulation "rabbit hole", and per-recommendation trade-off radars | TypeScript Recommender |
| Project | What it does | Stack |
|---|---|---|
| Credit Default Prediction | Credit-decision model on a lending dataset — leakage-aware, honest accuracy ceiling, best result from a diverse model stack | Python scikit-learn |
| Sabadell Capstone | Master's capstone with Banco Sabadell — applied analytics on a real banking problem | Python Analytics |
➕ More projects (AI experiments & data viz)
| Project | What it does | Stack |
|---|---|---|
| Boardroom Simulator | AI-driven boardroom decision simulator | Python AI |
| Hotel Booking AI | Predictive analytics for booking cancellations | Python ML |
| Hamlet Data Viz | NLP + visual analysis of Hamlet — word frequency, sentiment arcs, character networks | R ggplot2 NLP |
