Hi, my name is
I'm a machine-learning researcher and software engineer based in Accra, Ghana, finishing a B.Sc. in Computer Science at the University of Ghana. I fine-tune and deploy vision-language models, build NLP tools for low-resource languages, and engineer the backend platforms that hold it all together.
Get In TouchHello! I'm Benedict — a computer science student who fell in love with the why and how behind the models and frameworks everyone else just uses. That curiosity has taken me from debugging dependency conflicts in LoRA fine-tuning pipelines to publishing research on bias in content moderation systems.
My recent work sits at the intersection of machine learning and real impact: fine-tuning MedGemma to recognise tropical skin diseases on dark skin tones — a group dermatology AI largely underserves — and packaging Twi language data so other researchers can build for a low-resource Ghanaian language.
When I'm not training models, I lead backend engineering teams, ship payment-integrated platforms, run a student community that helps young builders turn ideas into real projects — and write about what I'm thinking through along the way.
Here are a few technologies I've been working with:
A full donation-management platform for causes across Africa. I led the entire backend in an agile team: REST APIs for donation tracking and cause management, JWT authentication, rate limiting, Celery background jobs, and Paystack payment-gateway integration supporting donations in multiple African countries. We later consolidated four microservices into a single deployable Django monolith with per-domain databases and a custom database router.
An installable Python package (pip install twicorpus) providing NLTK-style word lists and corpora for Twi — a low-resource Ghanaian language. Built to lower the barrier for NLP research in Akan, with open-source attribution to the GhanaNLP community dataset.
A smart service-request dispatcher for the University of Ghana campus, built jointly with classmates. Custom data structures (heaps, red-black trees, graphs), classic algorithms (Dijkstra, MST, knapsack), a simulation GUI and an empirical benchmarker — all from scratch in Java.
A prototype mobile app for Twi automatic speech recognition: realtime streaming transcription with English translation, dialect switching (Asante, Akuapem, Fante), transcript history with search and bookmarks, and export. The architecture is wired so a real inference backend can drop in.
An early-stage agent harness for locally hosted LLMs — designed around cross-model compatibility and an extensible marketplace of agent skills and tools.
A knowledge-engineering project: Prolog facts, rules and queries powering a mood-based music recommender, with a React Native frontend and Spotify playlist streaming. User feedback tunes the valence values that drive song selection.
AI agent that creates short animated videos from a brief; user approves each step and every video stays as editable files based on ICMs.
Hackathon project helping users find reliable financial literacy and investment platforms.
Business intelligence platform for Ghanaian small businesses; turns sales sheets, Mobile Money alerts and stock counts into clear numbers, honest insights and actionable answers. Features AI agent 'Lance' for instant business questions and supports 5 file formats including CSV and Excel.
First-author preprint investigating how content-moderation classifiers become more sensitive to certain topics during crisis periods, and how that heightened sensitivity turns into flagging bias. Contributes methods for measuring topic sensitivity at the intersection of NLP and algorithmic fairness.
Research conducted during my internship at DISAL: fine-tuning MedGemma 4B with LoRA for tropical skin diseases prevalent in Africa, with dataset curation, image-quality enhancement and systematic evaluation — deployed optimised for dark (Fitzpatrick V–VI) skin tones to reduce prediction bias.
I'm currently open to machine-learning engineering and research opportunities, internships, and interesting collaborations — especially at the intersection of AI and real-world impact. My inbox is always open, whether you have a role, a project, or just want to talk shop.
Say Hello