ISAIAH KOLAWOLE
ML/AI Engineer | Portfolio Profile

Email: gkola67@gmail.com
Phone: 08112324547 (+234 811 232 4547)
GitHub: https://github.com/kolawoleisaiah325
Portfolio: https://isaiah-kolawole-portfolio.vercel.app/

FOCUS
Building data pipelines, forecasting models, and AI applications that turn data into useful decisions, with a focus on reproducibility, data quality, and inspectable evidence.

FEATURED PROJECT: HEALTH PROGRAMME ANALYTICS
A reproducible Python pipeline for a synthetic facility network, combining reporting validation, a Streamlit dashboard, forecasting comparisons, and evidence-checked AI draft briefs.

- Distinguishes accepted counts, missing reports, and invalid submissions across 216 facility-months, retaining an audit trail.
- Compares three forecasting methods using a time-based holdout. Holt-Winters recorded MAE 27.38 doses and WAPE 3.61% across nine complete synthetic months.
- Supports a PostgreSQL schema and tested SQL views, alongside CSV outputs used by the dashboard.
- Uses local Ollama generation with citation and numeric checks, a deterministic fallback, and human review.

All data is synthetic. These figures do not establish real-world programme impact. The same holdout selected and evaluated the model; independent validation remains necessary.

TOOLS DEMONSTRATED IN THE PROJECT
Python, pandas, NumPy, scikit-learn, statsmodels, SQL/PostgreSQL, Streamlit, Ollama, Git/GitHub.
Project use is not a proficiency rating or professional experience claim.

EXPLORE
Case study: https://isaiah-kolawole-portfolio.vercel.app/case-study.html
Interactive demo: https://isaiah-kolawole-portfolio.vercel.app/demo.html
Source: https://github.com/kolawoleisaiah325/health-programme-analytics

FEATURED PROJECT: GITHUB CODE CHAT
Local AI codebase assistant with an offline bundled example, keyword retrieval, local embeddings, hybrid rank fusion, and Ollama-generated explanations.
- Reads public repositories at a pinned commit and preserves file/line references.
- Shows source excerpts behind answers; checks citation numbers and handles unavailable models.
- Includes 20 authored evaluation questions, recorded method comparisons, documented model errors, and automated tests.
Local engineering prototype. Evaluation on five authored files is not evidence of general repository performance or answer accuracy.
Case study and source search: https://isaiah-kolawole-portfolio.vercel.app/code-chat.html


FEATURED PROJECT: INSPECTAI
Visual anomaly inspection with frozen pretrained CNN features, normal-patch memory, ONNX CPU inference and FastAPI.
- Fits on 167 normal images and calibrates on 42 held-out normal images.
- Compares against a pixel baseline on all 83 official MVTec AD bottle test images.
- CNN image AUROC 0.9976; missed defects 2/63; normal images flagged 0/20. Both misses are documented.
- Includes a heatmap overlay and browser history with human review.
Noncommercial research prototype under CC BY-NC-SA 4.0. Scores are not probabilities; results do not establish performance on arbitrary uploads.
Tools demonstrated: PyTorch, NumPy, ONNX Runtime, FastAPI, Pillow.
Case study: https://isaiah-kolawole-portfolio.vercel.app/inspect-ai.html