14 Workflow Map¶
Use this page when you know what you want to do but not which document to open first.
flowchart TD
A["What do you want to change?"] --> B["Add or refine personal knowledge"]
A --> C["Ingest new source documents"]
A --> D["Tune model behavior"]
A --> E["Run or review training"]
A --> F["Evaluate quality"]
B --> B1["Write persona"]
B --> B2["Write domain knowledge"]
B --> B3["Snapshot knowledge state"]
B1 --> G["knowledge/persona.md"]
B2 --> H["knowledge/domains/"]
B3 --> I["knowledge/snapshots/README.md"]
C --> J["docs/02_data_ingestion.md"]
C --> K["docs/13_data_quality.md"]
D --> L["docs/09_guardrails.md"]
D --> M["docs/12_guardrail_boundary_cases.md"]
E --> N["docs/03_training.md"]
E --> O["docs/10_dataset_design_examples.md"]
F --> P["docs/07_evaluation.md"]
F --> Q["evaluation/cases/core_eval_cases.jsonl"]
I want to...¶
| Goal | Start here | Then go here |
|---|---|---|
| Add my own opinions and defaults | docs/11_personal_knowledge.md | knowledge/persona.md |
| Ingest local, NFS, email, or Linkwarden data | docs/02_data_ingestion.md | docs/13_data_quality.md |
| Build better SFT examples | docs/10_dataset_design_examples.md | data/training/examples/README.md |
| Make the model more or less restrictive | docs/09_guardrails.md | docs/12_guardrail_boundary_cases.md |
| Compare guarded versus original-model behavior | docs/09_guardrails.md | docs/07_evaluation.md |
| Run a first local LoRA job | docs/03_training.md | docs/10_dataset_design_examples.md |
| Review whether the model is actually good | docs/07_evaluation.md | evaluation/cases/README.md |
Recommended first-timer path¶
- read docs/00_stack_primer.md
- complete docs/01_installation.md
- write knowledge/persona.md
- fill in two or three files listed in knowledge/domains/README.md
- ingest your first sources
- create a small curated dataset
- run evaluation before and after LoRA training