AI & Data Science
From Raw Data to Real Impact
I build AI systems that solve problems people actually have, from computer vision pipelines to generative AI tools designed for real users. My work spans the full ML lifecycle: data preprocessing, model development, REST API integration, and communicating results clearly to non-technical audiences. I've taken projects from research to publication and from concept to deployment, with a focus on building things that work outside the lab, not just inside it. Currently seeking my first industry role in AI, ML, or Data Science, with Computer Vision and Generative AI as my core specializations, ready to hit the ground running from day one.
I'm an AI and Data Science graduate with a strong focus on Computer Vision and Generative AI. I build systems that don't just run in notebooks — they run in the real world.
My background spans deep learning, large language models, NLP, and full-stack AI integration. I approach problems with a researcher's rigor and ship solutions with a builder's pragmatism.
Outside of AI, I'm into football, digital design, and video editing — interests that keep my problem-solving creative and my eye for design sharp.
Undergoing an intensive, industry-aligned training programme that covers the complete data science and AI lifecycle, from data access and analysis to model development, Generative AI and full-stack deployment. The programme is project-based and combines theory and practical sessions with assignments, mini projects, a main project, machine tests and interview-focused training.
Evidence-based IT helpdesk assistant using Retrieval-Augmented Generation over a 30-document PDF knowledge base. Answers are grounded in retrieved source passages to reduce hallucination. Retrieval and generation quality are measured with RAGAS (faithfulness, answer relevancy, context precision and recall), targeting above 0.90 on the core metrics. Currently in active development, with a frontend and public deployment planned.
End-to-end hourly electricity-demand forecasting and tariff-aware cost simulation for small commercial buildings, built on the I-BLEND smart-meter dataset (4 buildings, 2013–2017). LightGBM cut forecast error from 20.5% to 8.7% WAPE against a seasonal-naive baseline, and MAE from 2.15 to 0.91 kWh. It beat Random Forest and SARIMAX across all buildings in leakage-safe, rolling-origin backtests. Served through a 7-endpoint FastAPI service and a 6-page React dashboard, containerised with Docker Compose and CI, with 96 passing tests.
An adaptive multi-modal writing system for neurodivergent learners, built as part of a 4-member research team. Designed specifically for individuals with dysgraphia, the platform delivers real-time AI-powered feedback across typing, handwriting, and speech input modes. It combines NLP-driven error detection, Google Cloud Vision OCR, and gamified learning exercises to support inclusive education. Submitted for publication in Springer Nature's Education and Information Technologies journal.
A non-invasive deep learning system that predicts blood group from fingerprint images. A LeNet-5 CNN classifies 8 blood groups (A±, B±, AB±, O±) and reached 73.08% test accuracy after 30 epochs. Fingerprints were resized to 224×224, normalised and augmented, then trained on an 80/20 split with the Adam optimizer. The model is deployed as a Gradio web app for instant predictions from an uploaded fingerprint. Published in the International Journal of Scientific Research in Engineering and Management (IJSREM, Vol. 9, Issue 4, 2025).
Open to entry-level roles in AI, Machine Learning, and Data Science. If you're building something interesting or have an opportunity that fits, reach out.