I've been building web scrapers for over three years, and today I deliver commercial datasets to the DataBoutique (DBQ) marketplace, covering fashion, luxury, real-estate, pharmacy and grocery sites across Europe, the Middle East and the US.

What I bring to a team:

• End-to-end ownership: target analysis → link discovery → scraping → enrichment → schema conversion → upload to S3.
• Anti-bot expertise: I'm comfortable working against Cloudflare, Akamai and other WAFs using modern stealth tooling.
• Operational discipline: every job runs on a schedule with Telegram monitoring, so failures are caught in minutes, not days.
• Scale: from catalog crawls with thousands of SKUs to a 2M+ post forum scraper with deduplicated MySQL storage.

Scraping & Automation

Data Engineering, ML & Storage

Infrastructure

Every project follows the same battle-tested structure, which keeps jobs resumable, debuggable and easy to hand over:

phase1  →  Discover brands / categories / listing links
phase2  →  Scrape product or listing data (resumable, deduplicated)
phase3  →  Enrich from detail pages (PDP)
phase4  →  Convert to the target DBQ schema (e.g. E0005r, E0001, GROCERY-PLP, REAL-ESTATE-BASIC)
phase5  →  Validate and upload to S3

🔗 Public code lives in my repositories below. Client-delivered scrapers are summarised here without proprietary data.

I treat scrapers like production services. Each one is scheduled on an Ubuntu VM and monitored through a shared Telegram bot:

• ⏰ Cron-scheduled runs via a standard run_scraper.sh wrapper
• 🔔 Instant alerts on job start, finish and failure through notify.sh
• 💬 Bot commands to check on jobs from my phone: /projectstatus, /projectlog, /help
• 🔁 One reusable setup across all active scrapers, so adding a new project takes minutes

MalariaScope is an AI-assisted tool that estimates malaria parasitemia from thin blood smear microscopy. Users upload several microscope fields from one slide, a YOLO model detects and classifies every red blood cell, and the app aggregates the counts into a single parasitemia percentage with CDC/WHO severity context. It was my capstone for the 3MTT Fellowship (Data Science, ML & AI track).

Judgment calls that went beyond "train a model on a dataset":

• Reframed the problem. Collapsed the dataset's 7 fine-grained classes into 2 (infected vs. not) after visually inspecting crops, and dropped leukocytes so they wouldn't corrupt the red-blood-cell denominator.
• Handled a ~28:1 class imbalance with copy-paste augmentation and a higher classification loss weight, prioritising recall on infected cells because missed infections are the costly error.
• Followed clinical counting protocol. Sums cell counts across all fields before computing one percentage, instead of averaging per-image percentages, which would misweight sparse and dense fields.
• Reported its own limits. The write-up documents the lower recall on parasitized cells and the need for cross-dataset validation.

My pharmacy training helps here: I understand the clinical problem the model is meant to serve, not just the metrics.

• B.Pharm, University of Lagos (2017 – 2024)
• 3MTT Fellowship, Data Science, ML & AI track (capstone: MalariaScope)
• Transitioned into data engineering and ML, bringing a detail-oriented, quality-first mindset from a regulated field.

I'm open to remote data engineering / web scraping roles and freelance data-extraction projects. If you need reliable, monitored data pipelines that survive bot protection, reach out on LinkedIn, Telegram or email.