ML Engineer and GenAI practitioner with hands-on experience building production-grade AI systems and full-stack web applications. Backed by 4 peer-reviewed publications and practical experience across AWS, Azure, and modern frontend frameworks.
Selected work

Full-stack food delivery platform built for a food service business. Features user authentication, a full menu system, custom order requests, and real-time order tracking from kitchen to door.

Professional multi-page website for a Ghanaian publishing house. Covers service listings, author submission flow, blog, and contact.
Guided German tax return wizard with live refund estimation, Lohnsteuerbescheinigung OCR, Werbungskosten optimizer, ELSTER XML export, and AI-powered result explanations. Built for employees and freelancers navigating Germany's complex §32a EStG tax system — in English and German.
Production RAG pipeline over 389 arXiv papers — 1,500 semantic chunks in ChromaDB, 0% zero-result rate, deployed with MLflow tracking and CI/CD to Hugging Face Spaces.
5-agent LangGraph StateGraph with typed state, conditional routing, and a Critic retry loop. 85% quality score at 7.6s end-to-end.
Modern, user-friendly site for Amdor Lodge, highlighting lodging options, charm, and booking essentials. Focused on clean UI/UX with dynamic elements to drive inquiries and position the business as a welcoming Kasoa destination.
Live weather sensor analytics dashboard — Open-Meteo Archive API, PostgreSQL, 10 analytical SQL queries, Streamlit + Plotly, CSV/Excel export. 6 European cities, 330 readings/refresh.
EfficientNet-B3 vision classifier for industrial casting defect detection — 99.7% accuracy, 100% recall on 715 test images, LIME explainability, Azure ML deployment pipeline, MLflow tracking.
LSTM Autoencoder for industrial sensor anomaly detection on NASA CMAPSS — AWS SageMaker endpoint, live Grafana + Prometheus monitoring, MLflow tracking, SHAP explainability, SageMaker Pipelines DAG.
XGBoost RUL forecasting for turbofan engine predictive maintenance — NASA CMAPSS, SHAP explainability, AWS SageMaker deployment, Grafana + Prometheus monitoring
LangChain-style agent for AI-powered engineering design — arXiv + Semantic Scholar knowledge base, ChromaDB, 4-tool pipeline (parse → retrieve → generate → evaluate), PDF + Word export, MLflow tracking, CI/CD via GitHub Actions, live on HF Spaces.
Built a full-stack personal finance tracker with a Next.js 15 App Router architecture, featuring per-user data isolation via Clerk authentication and a serverless Postgres database on Neon with Drizzle ORM. The application pipeline covers transaction ingestion with keyword-based auto-categorization, live multi-currency conversion via a cached exchange rate proxy, and an analytics endpoint aggregating spending into chart-ready payloads. Deployed to Vercel with CI/CD via GitHub push, with a PWA manifest and service worker enabling home screen installation on iOS and Android.
Sky-Sense is a hyperlocal weather experience that transforms its entire visual identity based on real-time conditions. Eight complete visual themes respond to the weather — sunny days get warm golden gradients and floating light particles, rainy evenings get deep blues and animated falling raindrops, stormy nights get dark purples and dramatic rainfall, clear nights get deep navy with twinkling stars. Every theme has its own colour palette, glassmorphism card styling, and canvas-rendered particle animation. When conditions change, Framer Motion cross-fades the entire interface in under a second. Beyond the visuals, the app surfaces genuinely useful information. A suggestion engine evaluates the combination of temperature, weather code, UV index, and wind speed to generate outfit recommendations and activity advice — telling users not just what the weather is, but what to do about it. Air quality data shows AQI alongside individual pollutant readings (PM2.5, PM10, NO₂, ozone) with plain-language health advice. A 24-hour scrollable hourly timeline and 7-day daily forecast complete the picture
A cross-platform mobile app for Germany's complex waste sorting system that scans any product barcode and instantly returns the correct bin assignment, solving a daily pain point for the 1.5 million immigrants who arrive in Germany each year without knowledge of local recycling rules
Client-side web calculator that gives employees and employers a complete real-time breakdown of Minijob and Midijob employment costs in Germany, incorporating all six employer flat-rate contribution categories, a live earnings limit tracker, Minijob-to-Midijob transition detection, and full EN/DE language support — updated for the 2025 earnings ceiling of €556/month
A tenant rights web application for the millions of Germany's renters incorporating a plain-English rights guide across six legal categories (BGB §535–§580a), a Mietpreisbremse rent legality checker comparing user rent against official Mietspiegel data for seven cities, and an AI-powered formal dispute letter generator producing simultaneous German and English versions across five dispute types including deposit challenges, repair requests, and eviction objections
Mobile guide to German bureaucracy covering 10 essential procedures (Anmeldung, Aufenthaltstitel, Steuernummer, Krankenversicherung, Gewerbeanmeldung and more) with step-by-step progress tracking, required document checklists, an AI assistant for follow-up questions, and automatic GPS-based office sorting that identifies the nearest relevant government offices on first launch using the Haversine formula — in English and German
RAG-powered resume tailoring tool — paste a job URL or description, upload your resume, and get a section-by-section fit score, ranked skill gaps with YouTube learning links per gap, and role-specific bullet rewrites using ChromaDB retrieval over the job description. Includes an ATS keyword density meter, a redline Word export with track-changes style markup, and a full CI/CD pipeline with GitHub Actions, Docker Hub, Railway, and Vercel.
Built a real-time anomaly detection engine using an ensemble of Isolation Forest, Local Outlier Factor (LOF), and an advanced TabNet Autoencoder (PyTorch) to identify fraudulent credit card transactions. Deployed via a modern Streamlit interface featuring live exchange rate integration to dynamically normalize international currency inputs for the ML models.
Technologies
Frontend
AI / ML
Backend
MLOps & Cloud
Databases
Education
Publications
Conferences
Contact
Let's work together
Open to ML Engineering and Full-Stack roles, and freelance projects. I respond within 24 hours.