SELECTED BUILDS
Projects with a practical reason to exist.
Personal and academic projects that demonstrate product thinking, secure engineering, data modelling and applied machine learning.
LOCAL / PRIVATE / HUMAN-VALIDATED
LOCAL-FIRST AI PRODUCT
CodeCVI Agent
A private, practical AI workflow designed as a real product rather than a disconnected collection of AI features.
The problemJob applications involve repetitive comparison, rewriting and document management, while CV data is highly personal and often sent to external AI services.
The approachUnify CV parsing, job matching, gap analysis, supported-claim validation, tailoring, cover letters, application tracking and privacy controls in a local-first architecture.
The current outcomeAn active Docker-based MVP with authentication, PostgreSQL, Redis, FastAPI services, a React workflow and local Qwen 2.5 generation through Ollama.
Why it matters professionallyDemonstrates privacy-aware architecture, responsible AI, workflow design and careful handling of sensitive personal information.
MODEL / QUERY / TRANSACT
RELATIONAL DATA MODEL
Employee Management SQL
A relational data project built around realistic employee, department, geography, consultancy, customer and sales questions.
The problemOperational reporting requires connected data structures and queries that can answer cross-departmental questions without duplicating or hard-coding information.
The approachDesign an 11-table company model and solve business stories using joins, subqueries, aggregation, grouping, variables and transactional updates.
The current outcomeNine core and four bonus reporting stories completed, including compensation comparisons, regional headcount, customer sales and transaction-safe changes.
Why it matters professionallyDemonstrates relational modelling, reporting logic and query patterns that transfer directly to people-data analysis and HR systems work.
DESIGN / CONTENT / DELIVERY
PRODUCTION WEB PLATFORM
Multilingual Portfolio Platform
The reusable engineering case study behind this website: identity, content, multilingual routes, management tools and hosting treated as one product.
The problemA professional portfolio becomes difficult to maintain when content, translations, certificates, deployment and visual presentation are handled separately.
The approachCombine a typed Next.js interface, three-language content, Sanity Studio, accessible navigation, responsive motion and Cloudflare-compatible output.
The current outcomeA production-oriented template with documented architecture, local content fallback, CMS schemas, social metadata, tests and a repeatable deployment path.
Why it matters professionallyDemonstrates production ownership, multilingual information architecture, accessibility, deployment and maintainable content systems.
TRAFFIC / CLASSIFY / RESPOND
CYBERSECURITY TOOLKIT
Android Malware Detection
A practical cybersecurity toolkit that moves malware classification from a research model into workflows a user can operate.
The problemAndroid malware can appear in stored datasets, captured network sessions or live device traffic, but each source requires a different analysis workflow.
The approachUse one LightGBM-based detection foundation across file analysis, PCAP classification and live USB-tethered traffic inspection, presented through Python GUI applications.
The current outcomeThree operational analysis paths: CSV/XLSX dataset scanning, PCAP inspection and real-time packet classification with Scapy and PyShark.
Why it matters professionallyDemonstrates analytical rigour, reproducible testing, model evaluation and security awareness.
COMPARE / EVALUATE / LEARN
FINAL-YEAR RESEARCH
ML/DL Algorithm Analysis
Final-year research comparing traditional machine learning, deep learning and hybrid ensembles for Android network-traffic classification.
The problemA high accuracy score alone does not show which model family offers the best balance of classification performance, generalisation and practical evaluation.
The approachEvaluate ML, DL and hybrid combinations using accuracy, precision, recall, F1, ROC-AUC, precision-recall curves and confusion matrices across multiple malware classes.
The current outcomeA documented comparative framework in which the strongest reported hybrid ML ensemble reached 99.3% accuracy and a 99.2% F1 score on the project dataset.
Why it matters professionallyDemonstrates evidence-led comparison, evaluation discipline and the ability to explain technical trade-offs.
RULES / SERVICE / ASSURANCE
TESTED JAVA SERVICE
Banking System
A service-focused Java project demonstrating how financial operations can be expressed as clear domain logic and verified behaviour.
The problemBanking actions must preserve predictable rules and state changes; small logic errors can produce incorrect balances or inconsistent customer outcomes.
The approachSeparate core banking responsibilities through object-oriented Java, manage the build with Maven and verify service behaviour with isolated JUnit and Mockito tests.
The current outcomeA documented personal banking system that demonstrates core functionality, service-layer reasoning and automated test coverage.
Why it matters professionallyDemonstrates dependable service logic, automated testing and disciplined handling of state-changing operations.