Awon Aziz — AI / MLOps engineer. Islamabad, Pakistan.
AI/MLOps engineer. LoRA fine-tuning on a hand-written autodiff with every layer proved against its reference, drift detection that is calibrated to report no, an evaluation warehouse that fails CI on a stale row, and scheduled automation that has run unattended for weeks.

Systems:
- From scratch to served: a LoRA fine-tune where nothing is taken on trust (2026, Shipped — 80 tests, every layer checked against its reference, INT8 served on CPU). Stack: Python, NumPy, PyTorch, peft, ONNX, ONNX Runtime, FastAPI, pytest. Most fine-tuning posts show a loss curve. This one shows a parity table. A reverse-mode autodiff written on NumPy, multi-head attention written out rather than called, LoRA on nn.Linear, then an ONNX export and a quantised FastAPI service — with each layer graded against the library it replaces and every number carrying an interval.
- Evaluation warehouse: six questions that are hard in a notebook and trivial in SQL (2026, Shipped — 141 tests, 5 migrations, a quality gate that fails CI). Stack: Python, DuckDB, SQLAlchemy, Pandas, Pydantic, FastAPI, Streamlit, Plotly. Every project here already writes evaluation artefacts — per-prediction arrays, serving reports, drift metrics. This puts them in a DuckDB star schema behind real migrations and answers the six questions that decide whether a model change was an improvement: LoRA against a full fine-tune, calibration by arm, INT8 against FP32, drift trend, error by document length, and which slices still lose.
- LLM drift monitor: a production shift, replayed, with the alerts it should raise (2026, Shipped — 184 tests, 4 CI jobs, runs offline in ~50s). Stack: Python, FastAPI, Ollama, qwen3:8b, sentence-transformers, SQLite, Streamlit, Prometheus, Docker. Most monitoring demos compare two histograms. This watches a real LLM application — a banking support assistant — across a scripted production shift and answers the question a platform team actually has: did the system get worse, and what should we do about it?
- Incident copilot: retrieval that knows when to distrust itself (2026, Advisory only — three builds, 38 tests on the current one). Stack: Python, Chroma, CrewAI, FastAPI, Kubernetes, scikit-learn. An anomaly detector tells you a metric is strange. It does not tell you why, and it has no memory. This searches a postmortem knowledge base for incidents that resembled this one and puts a drafted root-cause hypothesis in front of a human.
- Job funnel: eighteen boards, every twenty minutes, no scraping (2026, Running unattended since 16 Aug 2026 — 587 automated commits). Stack: Python, GitHub Actions, GitHub Pages, requests. Nearly every company careers page is a thin client over an applicant tracking system, and those systems publish the listings as JSON at a public endpoint. A scheduled workflow reads those endpoints, filters, dedupes against the previous run and commits the result to a static page.
- Model lifecycle: drift, retrain, and a promotion that can be refused (2026, Shipped — 43 tests, 3 drift measures, synthetic data). Stack: Python, FastAPI, MLflow, Evidently AI, SciPy, Streamlit, Docker. A model that passes every test and still quietly becomes wrong is the failure mode CI cannot see. The service stays healthy, no test fails, and the predictions stop meaning anything. This is the machinery for catching that.
- AI pair engineer: four agents, and a budget on what each is told (2026, Shipped — 4 stages, 20 tests, never executes your code). Stack: Python, Pydantic, Streamlit, OpenRouter. A four-stage review pipeline that runs before a pull request exists: analyse, generate tests, refactor, then compare the original against the refactor and check behaviour survived. It stops at the report.

Repositories:
- https://github.com/AwonAziz/from-scratch-to-served
- https://github.com/AwonAziz/eval-analytics-platform
- https://github.com/AwonAziz/llm-drift-monitor
- https://github.com/AwonAziz/Hybrid-retrieval
- https://github.com/AwonAziz/agentic-incident-copilot
- https://github.com/AwonAziz/ai-incident-response-system
- https://github.com/AwonAziz/cleanjobfunnel
- https://awonaziz.github.io/cleanjobfunnel/
- https://github.com/AwonAziz/ml-lifecycle-platform
- https://github.com/AwonAziz/AI-Pair-Engineer

Contact: awonaziz786@gmail.com