Yes — most of Week 1 is done. I would consider the core Week-1 product complete because the repo already has:
apps/web, packages/db, packages/shared, packages/ai setup.Document, ExtractionRun, ExtractionEvent, with statuses UPLOADED → EXTRACTING → VALIDATING → COMPLETED / NEEDS_REVIEW / FAILED.rawModelOutput, extractedJson, confidenceJson, model, and promptVersion.validationJson, missingFieldsJson, and final status./review queue foundation reading NEEDS_REVIEW runs.Yes. Definitely.
Before starting Week 2, spend a short session creating a Week-1 demo. This is what turns the work from “I built some routes” into “I shipped a product slice.”
Your Week-1 demo should show:
1. Upload valid claim PDF/email
2. Run extraction
3. Show structured JSON
4. Run validation
5. Show COMPLETED
Then:
6. Upload incomplete/repair-estimate/theft claim
7. Run extraction
8. Run validation
9. Show NEEDS_REVIEW
10. Open /review
11. Open run detail timeline
Add this to README:
## Week 1 Demo
### Demo flow
- Valid claim → COMPLETED
- Incomplete claim → NEEDS_REVIEW
- Timeline proves each workflow step
### Known limitations
- No real ReviewTask table yet
- No human edit/approve/reject yet
- Extraction is still manually triggered
- No durable background worker yet
Do not spend a full day polishing Week 1. Just create proof: screenshots, a short video, and a README demo script.
Your saved 8-week direction is:
Week 1: Document Intake Reviewer
Week 2: Durable Workflow + Human Review
Week 3: RAG over policy documents with citations
Week 4: Agent/tool routing workflow
Week 5: Memory from reviewer corrections
Week 6: Gateway, governance, observability, evals
Week 7: Terminal/repo assistant or company-specific proof-of-work
Week 8: Fine-tuning decision lab + final hiring demo