Faris Hambo

A personal architecture prototype for an agent-assisted software delivery lifecycle.

I designed a queue-based flow that moves work from requirements through implementation, testing, security review, deployment, and post-deployment incident feedback. The repository includes an initial Python harness that validates task contracts, loads task-specific skills and context, routes model calls, evaluates acceptance criteria, and applies hard guardrails around protected branches, CI failures, and production approvals. I built the project to explore auditable, human-gated agent workflows and a clear separation between LLM judgment and deterministic execution. It is an early-stage technical prototype rather than a production deployment.

Built a side-by-side machine learning pipeline showcase using Ordeq and Kedro to compare modular pipeline design, execution, testing, and visualization patterns.

The project includes a classical ML workflow covering data preparation, feature engineering, train/test splitting, model training, evaluation, persisted metrics, and visual outputs. It also includes an LLM text-classification workflow with dataset preparation, Hugging Face model loading, inference, prediction analysis, and unit tests. I implemented reusable data catalogs and datasets, custom model IO, execution timing and logging hooks, command-line entrypoints, Mermaid/Kedro visualization, and isolated tests for pipeline nodes. The project was built as a technical evaluation of maintainable ML pipeline patterns rather than as a distributed production orchestrator.