Pipeline Core
Python CLIs, OmegaConf configs, stage manager, resource monitor, and metadata I/O.
CPU-safe demo for a GPU-heavy ML workflow
Turn 2D animation footage into LoRA-ready character datasets.
This public demo shows the reviewable surface of the pipeline: frame extraction, YOLO tracking, ToonOut-style segmentation, identity clustering, pose conditioning, captioned datasets, and LoRA training snapshots.
Demo ready
mock-safe mode
7/7
Stages
7,924
Rows
339
Assets
Product results
Public artifacts designed for screenshots and walkthroughs
A public, mock-safe demo layer for a GPU-heavy character LoRA pipeline.
End-to-end flow
Inspectable 2D animation pipeline stages
The public demo uses deterministic synthetic data, so reviewers can evaluate the architecture without private footage, model files, API keys, or a CUDA workstation.
Demo scenarios
What to show in an interview
Deliverables
What this showcase proves
Screenshots and video
Ready-to-share review assets
These assets are committed with the static site and mirrored into the main portfolio media gallery.
Architecture
Config-driven ML workflow, static public demo
Vision Layer
YOLO/ByteTrack tracking, ToonOut-style masks, DWpose previews, and identity clustering.
Training Layer
Captioned LoRA datasets, kohya/diffusers configs, checkpoint metrics, and prompt tests.
Demo Layer
Static manifest JSON, synthetic screenshots, WebM video, GitHub Pages, and Docker/Nginx.
Runbook
Local demo commands
pip install -r requirements/core.txt -r requirements/dev.txtpython scripts/demo/run_demo_pipeline.py --skip-pipelinepython -m pytest tests/demo tests/test_end_to_end_pipeline.py -qpython -m http.server 8080 -d portfolio-web