ML workflows / Private experiment
Sakura
Coordinate data preparation, model training and reporting through an agent-driven workflow.
- My contribution
- I built the staged ML tools and orchestration described in the Sakura project note.
- Tools & methods
- Python · Streamlit · SQLite · Ollama
The problem
What this project investigates
An ML workflow repeats data loading, profiling, feature preparation, training and reporting. This project explores having an agent select and coordinate those steps.
The build
What I built
The documented design separates deterministic ML tools from the language model that chooses what to run. It includes profiling, training, evaluation and report generation.
Technical choices
Decisions behind the build
Separate decisions from execution
The agent selects tools; the tools perform the ML operations.
Test the pieces and the sequence
The note describes tool tests and end-to-end orchestration checks, but their results are not included here.
Current status
What the evidence shows
The repository provides a screenshot and a development account, not the private implementation. The note and project listing disagree on the number of development phases. Autonomous reliability and sandbox security are not established by this material.
Further work & evidence gaps
A reproducible sample run with artifacts and failure cases would make the current scope easier to assess.