Work
Case studies covering the problem, the constraints, and how each system was put together.
6 projects shown
Completed
Solo Developer
CLI tool that vectorizes entire codebases and exposes them via Model Context Protocol (MCP), enabling AI agents to semantically navigate large projects.
Completed
Solo Developer
NLP spam detection system using zero-shot classification (BART-MNLI) with rule-based heuristics and risk scoring — allow / shadow-review / block actions.
Completed
Solo Developer
A command-line tool that detects whether an image is AI-generated or real. It combines 10 detection methods, from metadata checks to frequency and texture analysis, into a weighted multi-signal score with a confidence level.
Completed
Solo Developer
A semantic matching microservice used in the jobs.af applicant tracking system. A gRPC service scores applicants against a job with Sentence-Transformers (falling back to TF-IDF) and returns an explainable breakdown: skills, experience, education, functional area, location and languages.
Completed
Solo Developer
The Python prototype of the jobs.af job recommender, now used inside the core jobs.af platform: a FastAPI service that embeds jobs and applicants with a multilingual sentence transformer, searches them with pgvector, and scores matches on skills, experience and detected job domain in about 780 ms for 200 recommendations.
Completed
Solo Developer
A movie recommendation system with two modes: content-based suggestions from genre similarity, and personalised picks from a user’s favourite genres. Data lives in PostgreSQL, run with Docker.
