Skip to content
collins madu.
Work archive

The systems, and the
thinking behind them.

Business systems, independent demonstrations,
and the work in between.

23 case studies · 73 projects in all

Compact archive
Demonstration

LeaveDesk

A working demo of a leave and approval portal: an escalation ladder derived from policy and carrying its own reason, an approvals queue that separates your turn from someone else's, a permission matrix rendered from the rules the server enforces, and impersonation that names the real actor.

Architect & sole engineer
2026
Demonstration

OpsPilot

A portfolio piece demonstrating multi-tenant SaaS architecture: a workspace-isolated data model, a server-enforced Owner/Admin/Member RBAC matrix, server-action workflows, and an 8-page client/project/task suite.

Architect & sole engineer
2026
Case study

Recko

Reconciliation engine with reference and amount/date tolerance matching and automatic break categorisation, plus a tiered/percent/fixed commission engine with refund reversal and a live ops dashboard.

Architect & sole engineer
2026
Demonstration

OfferGrade

A working demo of supplier-offer intelligence: ingests price lists in four formats and seven languages, normalises them, gates out anything it cannot identify with confidence, and scores the rest BUY / NEGOTIATE / REVIEW / IGNORE with per-line reasoning.

Architect & sole engineer
2026
Demonstration

FlowForge AI

A portfolio piece demonstrating LLM automation built to production shape: structured JSON extraction with fail-closed Zod validation, confidence-based routing to a human review queue, prompt versioning, and per-run step logging.

Architect & sole engineer
2026
Case study

AI Call Analysis

Async upload-to-analytics pipeline with a resilient Whisper → faster-whisper → mock transcription chain, GPT-4o-mini QA scoring against weighted parameters with a heuristic offline fallback, and PDF/CSV export.

Architect & sole engineer
2026
Case study

AI Marketing RAG Assistant

Hybrid BM25 + semantic retrieval over ChromaDB, grounded generation with hallucination and brand-voice quality scoring, and a cron/event trigger engine dispatching campaigns to email, Slack and webhooks.

Architect & sole engineer
2026
Case study

PE Deal Sourcing & Qualification

BANT-based lead qualification via LLM with a deterministic heuristic fallback, a multi-stage deal pipeline state machine, and real-time funnel and conversion analytics.

Architect & sole engineer
2026
Case study

Supplier Offer Ingestion & Normalization

Two-phase Python pipeline: MSAL device-flow ingestion of price-list emails from Microsoft Graph, then a 2,287-line heuristic normalizer handling header detection, headerless tables, and net-vs-RRP price classification.

Data & Systems Engineer
2026
Demonstration

DataHarbor

A portfolio piece demonstrating a scraping and data-quality pipeline: a dependency-free Python engine (fetch, clean, validate, dedupe, export) bridged into a Next.js operator dashboard with a 9-screen workflow and live data-quality analytics.

Architect & sole engineer
2026
Demonstration

Style DNA

A working demo of visual fashion discovery: per-garment detection rather than whole-photo embedding, content-hash caching, an interpretable eight-dimension vector space, and a brand-diversified feed where every result explains why it surfaced.

Architect & sole engineer
2026
Case study

Artika Studio

AI art studio for interiors: room-photo analysis with server-side palette extraction and vision-model wall detection, artwork generation and recolouring, drag-to-place composition, and an exportable placement report.

Architect & sole engineer
2025–2026
Case study

Modern Abaya E-commerce

Storefront and admin dashboard with Paystack payments, an S3 image pipeline, drag-and-drop product reordering, and real-time cart state.

Full-Stack Developer
2025

More from the archive

Smaller builds and focused experiments.

50 projects
Full-Stack2025–2026

Personal Portfolio & Engineering Journal

This site. Next.js 15 App Router with an MDX case-study engine, a MongoDB-backed contact pipeline with email notification, view-transition routing, and full SEO infrastructure.

Designer & Engineer
Next.js 15React 19TypeScriptTailwind CSSMDXMongoDBFramer Motion
collinsmadu.com
Client work2026

SAM3 Text-Prompted Segmentation Pipeline

Fine-tuned SAM3 inference with a tiled scheme — feathered overlap and max-confidence compositing — that segments high-resolution images without seam artifacts, packaged as a client-deliverable notebook workflow.

ML Engineer
PythonSAM3PyTorchJupyter
Client work2026

Finance Automation — Blotter & FX Treasury

A Node blotter-consolidation service (file watcher, column mapping, per-source dedup, timezone-aware weekly email with failure alerts) plus a Python FX treasury pipeline with Prophet forecasting and MAE/MAPE backtesting.

Automation Engineer
Node.jsPythonProphetMongoDBFastAPIReactRecharts
Mobile2026

Cooperant

A pre-launch managed-savings product for Nigerian corporate workers: native SwiftUI and Jetpack Compose member apps backed by a Spring Boot platform with employer and operations portals, a balanced ledger, payroll reconciliation, independent payout approvals, and live payment, identity, messaging and push-provider adapters.

Product Architect & sole engineer
SwiftUIKotlinJetpack ComposeJava 21Spring BootPostgreSQLFlywayDocker
Client work2025

Employee Leave & Approval Portal

Leave management with multi-level approval workflows, RBAC including super-admin impersonation and audit logging, Azure AD SSO, reimbursables, and SendGrid/Teams notifications.

Full-Stack Engineer
TypeScriptReact 19Express 5MongoDBAzure AD / MSALMicrosoft GraphSendGridRecharts
Full-Stack2024–2025

Play & Win Platform

White-labelled promotional gaming platform built solo and running on in-store kiosk tablets: one product delivered to two targets — a React web build for the kiosks and an Expo/React Native Android build for customers’ own phones. Arabic RTL throughout, WhatsApp OTP registration with international country selection, and games including a Skia-animated Spin the Wheel, Catch the Item, Memory Match, a gift-box reveal and a timed Tic Tac Toe against an AI opponent — each issuing redeemable prize codes.

Full-Stack Developer
ReactExpoReact NativeSkiaExpressMongoDB@zxingNetlify
AI Systems2025

Private LLM / RAG Document System

Secure retrieval-augmented generation with document ingestion, vector indexing, auth-aware query APIs, and structured JSON outputs for enterprise use.

AI Engineer
PythonLangChainVector DBFastAPIOpenAI
AI Systems2025

YOLO Object Detection API

Custom YOLO model trained with Ultralytics, exported to ONNX, and served as a real-time inference API behind FastAPI and Docker.

ML Engineer
PythonUltralyticsONNXFastAPIDocker
Client work2025

Scam Detector API

Perceptual-hash (pHash) duplicate-image fraud detection for marketplace listings, exposed as a batch /check_scam endpoint and packaged as a Docker image for an ingestion pipeline.

Backend / ML Engineer
FastAPIMongoDBpHashDocker
AI Systems2025

GreetIQ — AI Voice Receptionist

Voice-receptionist backend ingesting Vapi call webhooks with HMAC-SHA256 signature verification, classifying leads, and serving an admin API behind a React dashboard.

Backend Engineer
Node.jsExpressMongoDBVapiReact 19
AI Systems2025

Dealer Detector

Rule-based NLP classifier labelling car-marketplace sellers as dealer vs. private flipper across 30+ weighted signal flags, with multi-currency price normalization and a batch-CSV service.

Python / NLP Engineer
PythonFastAPIpandasRule Engine
AI Systems2025

Crypto Price Predictor

End-to-end ML price prediction: multi-source market scraping, engineered technical indicators, a RandomForest next-day movement classifier, and a served API with Streamlit and Next.js frontends.

ML Engineer
Pythonscikit-learnFastAPIStreamlitNext.jsRailway
Client work2025

Outlook Attachment Ingestion & Price Comparison

Multi-phase pipeline ingesting supplier price-list attachments from Outlook via Graph/MSAL, normalizing with fuzzy matching and EAN validation, and generating a cross-supplier comparison.

Data Engineer
PythonMicrosoft GraphMSALSQLAlchemypandas
Automation2025

Hair Price Extractor

OCR and vision-model extraction turning messy supplier price-sheet images into a validated, format-preserving Excel workbook, with per-supplier YAML rules, anomaly flagging and auto-generated review sheets.

Automation Engineer
PythonOpenAI VisionPydanticopenpyxlYAML
Scraping & Data2025

UK Baby & Toddler Class Aggregator

Modular Node pipeline aggregating class timetables from 13+ providers via reverse-engineered JSON APIs, headless-browser scraping and Tesseract OCR fallback, geocoded and deduplicated into a national directory.

Data Engineer
Node.jsPuppeteerPlaywrightCheerioTesseract.jsGoogle Places APISupabase
Scraping & Data2025

Franchise Scraper

Multi-source Node scraper aggregating UK franchise class listings from JSON APIs and HTML, geocoding via Google/OpenStreetMap with caching and progressive fallback, normalized to a unified 30-field schema.

Data Engineer
Node.jsPuppeteerCheerioGoogle Places APIOpenStreetMapSupabase
Scraping & Data2025

WooCommerce / Shopify Product Scraper

Generalized YAML-configurable scraper with sitemap and collection discovery, optional Playwright rendering, UA rotation and backoff retries, exporting normalized product import CSVs.

Data Engineer
PythonPlaywrightYAMLpandas
Scraping & Data2025

URL & Email Extraction Crawler

Configurable depth-N crawler and email harvester with same-domain scoping, skip-trap filters and a clean modular architecture.

Python Engineer
PythonrequestsBeautifulSoup
Scraping & Data2025

Dermstore / Eminence Review Scraper

Multi-site scraper extracting per-product review counts across four skincare retailers on three different review platforms — handling shadow-DOM widgets, lazy loading, concurrency and retry passes.

Data Engineer
Node.jsPuppeteer-extraPlaywrightp-queueExcelJS
Scraping & Data2025

Alibaba Supplier Scraper

Playwright-driven supplier scraper with CAPTCHA-aware session persistence and cookie injection, parsing company profiles into a structured multi-sheet Excel workbook.

Data Engineer
PythonPlaywrightopenpyxl
Scraping & Data2025

Family-Friendly Lead Scraper

Google Maps lead scraper filtering businesses by rating, then crawling their sites to extract qualified contact emails into per-city CSVs.

Data Engineer
Node.jsPuppeteer-stealthCheerio
Scraping & Data2025

Marrakech Hotels Lead Pipeline

Four-stage pipeline scraping Booking.com listings, enriching via Google Maps, and producing a cleaned, deduplicated leads CSV.

Data Engineer
PythonPlaywrightpandasGoogle Maps API
Scraping & Data2025

Jimshooter Site Archiver

Polite same-domain crawler and multi-stage export pipeline that archived and indexed an entire site's pages and download links.

Python Engineer
PythonrequestsBeautifulSoupJSONL
Client work2025

Shopify → WooCommerce Migration Scraper

Product migration tool consuming the Shopify products.json API and generating WooCommerce import and ACF variant CSVs for a store migration.

Data Engineer
PythonShopify APIpandas
Scraping & Data2025

AI E-commerce Scraper

Amazon scraping service pairing a Selenium anti-detection bot with a FastAPI + MongoDB backend and a Streamlit control UI.

Full-Stack / Data Engineer
PythonSeleniumFastAPIMongoDBStreamlit
Client work2025

Leave Portal Documentation Site

Live Docusaurus documentation site for an internal HR product: ~550 lines of task-based MDX guides with 70+ annotated screenshots, a customised React theme, and a Netlify pipeline with security headers.

Technical Writer & Engineer
Docusaurus v3ReactMDXNetlify
ERP & Internal2024–2025

Internal IT Operations & Security

Endpoint and account management, biometric integration, access reviews, and internal security practice for a small firm.

IT Operations
Azure ADMicrosoft 365BiometricsSOP Design
Creative Tooling2025

Artwork-on-Wall AI Preview

R&D exploration for realistic AI image placement on walls — ControlNet, MiDaS depth estimation and Segment Anything for room-aware compositing, prototyped in Gradio and Streamlit.

AI Engineer
PythonControlNetMiDaSSegment AnythingGradioFastAPI
Full-Stack2024

News Quote — Strapi CMS Backend

Headless CMS backend for a political-quotes and fact-check platform: ~13 interrelated content types and a moderation lifecycle that auto-publishes community submissions with deduplicated, verified citations.

Backend Engineer
Strapi v4Node.jsGraphQLRESTPostgreSQLRailway
Full-Stack2024

Hilance / Roboti Landing Page

Full-stack marketing site with a contact-form email relay and a deduplicated newsletter subscription flow with automated confirmation emails.

Full-Stack Developer
ReactTailwind CSSExpressMongoDBNodemailerDockerRailway
Full-Stack2024

Next.js E-commerce Enhancements

Cart badge sync, real-time state, admin order visibility, and a shared component system across a live storefront.

Frontend Engineer
Next.jsReactTypeScriptTailwind CSS
AI Systems2024

AI Market Insights Dashboard

Classical data processing combined with LLM narrative generation to produce analyst-quality market reports with interactive visualisations.

Data / AI Engineer
PythonFastAPIPandasscikit-learnOpenAIReact
AI Systems2024

Political Quote Extraction System

Daily pipeline: scrape news sites, extract politician quotes via LLM, entity-match to a CMS, deduplicate, and surface a review interface.

AI / Automation Engineer
PythonOpenAI GPTPlaywrightStrapiNode.js
Automation2024

Lead Enrichment → CRM Pipeline

Automated lead sourcing, enrichment and deduplicated push into a sales CRM with field mapping and error reporting.

Automation Engineer
PythonBreakcold APIPandasREST APIs
Automation2024

Shopify Product Description Automation

Bulk LLM generation of product copy written back to a live Shopify catalogue with tone control and rollback.

Automation Engineer
PythonOpenAI GPTShopify API
Automation2024

Social Media Poster Automation

Scheduled multi-platform content generation and publishing with per-channel formatting.

Automation Engineer
PythonOpenAI GPTSocial APIsCron
Automation2024

Blog Post Automation

Topic-to-published-article pipeline with outline generation, drafting, image selection and CMS publishing.

Automation Engineer
PythonOpenAI GPTCMS APIs
Scraping & Data2024

Etsy Review Scraper

Review extraction with pagination handling, rate limiting, and structured CSV export.

Data Engineer
PythonPlaywrightPandas
Scraping & Data2024

Google Maps Business Scraper

Location-scoped business extraction with contact enrichment and per-city CSV output.

Data Engineer
PythonPlaywrightPandas
Scraping & Data2024

News Quote — Remix Admin & Scraper

Remix admin tool scraping news articles with a stealth Puppeteer browser and extracting politician-attributed quotes via GPT-4o, persisting deduplicated results to a CMS on a daily cron.

Full-Stack Engineer
RemixTypeScriptPuppeteerOpenAIStrapiDocker
Scraping & Data2024

Scraper Admin Tool

React + Express/Puppeteer admin that scrapes article URLs and associates extracted data with entities in a CMS, with debounced search and live scrape logs.

Full-Stack Developer
ReactExpressPuppeteerStrapi
Scraping & Data2024

Inview.nl Authenticated Scraping

Session-authenticated extraction behind a login wall with token refresh and structured export.

Data Engineer
PythonPlaywrightRequests
Scraping & Data2024

Buku.app Book OCR Pipeline

Authenticated login, page-by-page capture across PDF/canvas/XHTML rendering, OCR fallback, and structured text + metadata output.

Data Engineer
PythonPlaywrightOCRPDF Processing
Mobile2024

Android Mobile Expansion

Native Android delivery of an existing product surface, with release signing, bundling and store submission.

Mobile Developer
AndroidJava/KotlinGradlePlay Console
ERP & Internal2024

ERPNext Setup & Operations

Deployment, module configuration, user and role setup, and operational rollout for an internal ERP.

ERP Engineer
ERPNextFrappePythonMariaDBLinux
ERP & Internal2024

Odoo Evaluation & Workflow Mapping

Business process mapping, module fit assessment, and a migration recommendation against existing workflows.

Systems Analyst
OdooProcess MappingBRD
Creative Tooling2024

Artika Price Builder

Configurable pricing engine for print products across sizes, framing options and materials, with margin rules.

Full-Stack Developer
TypeScriptReactPricing Rules
Creative Tooling2024

Artcha / Artika Workflow Automation

Metadata generation, listing preparation and publishing automation for a digital art catalogue.

Automation Engineer
PythonOpenAI GPTREST APIs
Creative Tooling2024

Artika GPT Workspace & SOPs

Structured workspace setup with folder organization, SOPs, templates and tone consistency for a creative brand.

Process / Knowledge Engineer
OpenAI GPTDocumentationSOP Design
Full-Stack2023

Bulk Image Upload & Share Platform

Bulk image uploading, gallery sharing, and scalable object storage with signed access.

Full-Stack Developer
ReactNode.jsExpressAWS S3MongoDBNetlify
Working together

Something worth
building?

Tell me about the workflow, the team,
or the problem you keep working around.

Let’s talk
What do you charge?

€70 an hour as a baseline, covering architecture, backend, frontend, deployment and documentation. Where a project can be scoped properly up front, I prefer to quote a fixed price.

When could you start?

Usually within a week, sometimes the next day. The variable is how much research a project needs before I can scope it responsibly. Most engagements run four to twelve weeks; some become ongoing work.

Can you take over an existing codebase?

Regularly. Several projects here began as existing repositories. The case studies explain what I inherited, what I changed, and the role that was mine.

How do you handle confidential work?

Client systems are described by sector, never by name. I share the problem, my contribution and the stack without exposing client data or private source. Any permitted demo captures use invented data. The artwork here is conceptual.

What do I get at the end?

A deployed system, the repository, and documentation written for whoever runs it next: architecture notes, a runbook, and operator SOPs.