AI/ML Data Engineer @ MEEAMI Technologies — Python, SQL & PyTorch; RAG/LLM systems, built in public one commit at a time.
Data Engineering · Data Science / ML · Data Analytics | Atlanta metro, remote anywhere in the US | available in two weeks
- 💼 AI/ML Data Engineer at MEEAMI Technologies — LLMs, RAG & conversational AI
- 🔭 I ship small, runnable, documented projects — real code with real logs, no vaporware
- 📅 One Python exercise every day in python-daily — data pipelines, ML patterns, LLM building blocks
- 🧠 Currently deep in PyTorch — latest: a neural network that scores 98/100 on the FizzBuzz interview
- 📊 Day job muscle: SQL, data quality, and testing — I find the row that breaks the report
- 🎓 Indiana Wesleyan University · certifications from Microsoft, Google Cloud, MathWorks & Coursera below
Data engineering & SQL
| Project | What it proves |
|---|---|
| 🧾 claims-data-quality | Config-driven data-quality checks for healthcare CSVs: orphan keys, duplicates, range violations, code formats, date logic. Stdlib only |
| 🗄️ sql-drill | A fresh randomized healthcare database every session, graded questions, your result set checked against hidden reference queries |
| 🧰 data-file-toolkit | CSV/JSON conversion and validation CLI, stdlib only, tested |
| 📅 python-daily | One runnable exercise a day: data pipelines, data quality, sqlite3, log parsing, PyTorch drills |
Data science & evaluation rigour
| Project | What it proves |
|---|---|
| 👶 birthweight-leakage-pytorch | Target leakage caught: one column takes the model from 62.6% to a perfect 100% / AUC 1.000 |
| 🩺 maternal-risk-pytorch | 55% duplicate rows. Naive split shows +37.9 points of "skill", honest dedup shows +8.9 |
| 🚫 abuse-detection-shortcut | Mask 25 tokens and the classifier drops below its own majority baseline. Frequency-matched control included |
| 🎗️ cervical-screening-pytorch | The baseline beat my model, 93.6% to 87%. Published as-is rather than tuned until it looked better |
ML, LLM & retrieval
| Project | What it proves |
|---|---|
| 📚 hands-on-llm-verified | Reproduces O'Reilly's Hands-On LLMs against 2026 libraries, 42/42 verified, and finds a fix that silently drops weighted F1 from 0.80 to 0.33 |
| ⚡ sentiment-analysis-api | FastAPI service, TF-IDF + logistic regression, 84.2% held-out, 23 offline tests, Dockerized, documents its own out-of-vocabulary failure mode |
| 🔎 tiny-rag-pytorch | The R in RAG in ~150 lines of pure PyTorch: chunk, embed, cosine retrieve. No LangChain, no vector DB |
| 🔍 bm25-search | Zero-dependency BM25 engine in pure Python: inverted index, Lucene-style IDF, CLI, 30 pytest tests, CI on 3.9–3.12 |
| Certification | Issuer | Issued |
|---|---|---|
| Annotate and Analyze Objects for Vision | Coursera | Mar 2026 |
| Introduction to Generative AI for Developers with Copilot | Microsoft | Feb 2026 |
| Introduction to Image Processing | MathWorks | Feb 2026 |
| Model Training & Evaluation | Coursera | Feb 2026 |
| NVIDIA: Fundamentals of Machine Learning | Whizlabs | Feb 2026 |
| Python Programming Fundamentals | Microsoft | Feb 2026 |
| Introduction to SQL for BigQuery and Cloud SQL | Google Cloud | Jun 2025 |
Credential IDs are listed on my LinkedIn.