5-Day Sprint Plan

Phase 4 Prototype Execution Roadmap
Use Case: Unstructured Data Extraction & Quality Improvement
Duration: 5 business days (Mon-Fri, March 10-14, 2026)
Team: 1.5 FTE (Architect 0.5 + Engineer 1.0)
Insurance Client
Use Case 1
Confidential

Sprint Goals

Execution Focus
Day 1 (Monday) — Foundation & UI Scaffold

Goals

  • Set up HTML/JS project structure
  • Build document upload interface (drag-and-drop)
  • Create extraction results display (JSON → formatted UI)
  • Start quality dashboard mockup

Deliverables

  • Uploadable file form (accepts PDF/images)
  • Sample extraction results JSON (hardcoded for now)
  • Basic results display showing extracted fields
  • Confidence score per field (0-100%)

Team Effort

Engineer: Full build (HTML/CSS/JS structure)

Architect: Review, quality gates design

Success Criteria:
✓ Upload form working
✓ Results display functional
✓ Basic styling complete
Day 2 (Tuesday) — Extraction Engine & Quality Logic

Goals

  • Integrate LLM (Claude/GPT-4 or Azure DI) for mock extraction
  • Implement quality scoring algorithm
  • Add anomaly detection (flag out-of-range values)
  • Build sample data pipeline (50 documents)

Deliverables

  • LLM integration (API calls for document extraction)
  • Quality scoring function (field confidence, missing fields, standardization)
  • Anomaly detection rules (policy start > end date, coverage limits)
  • 50 sample documents indexed and ready for processing

Team Effort

Engineer: LLM integration + quality logic

Architect: Data quality rules, extraction schema design

Success Criteria:
✓ Extraction working end-to-end
✓ Quality scores visible
✓ Sample documents loadable
Day 3 (Wednesday) — Dual Personas & Side-by-Side

Goals

  • Build Claims Adjuster dashboard (focus: coverage, claim eligibility)
  • Build Underwriter dashboard (focus: risk assessment, medical details)
  • Implement side-by-side document viewer + extracted data
  • Add highlighting (source text → extracted field mapping)

Deliverables

  • Claims Adjuster tab: Coverage details, exclusions, beneficiary, claim limits
  • Underwriter tab: Risk factors, medical history, underwriting flags
  • Side-by-side PDF viewer (left) + extracted JSON (right)
  • Text highlighting: Original document text linked to extracted field

Team Effort

Engineer: UI for personas, PDF viewer integration, highlighting logic

Architect: Workflow validation, UX review

Success Criteria:
✓ Both personas fully functional
✓ PDF viewer working
✓ Highlighting interactive
Day 4 (Thursday) — Human Correction & Polish

Goals

  • Build edit/correction interface (allow humans to fix misextractions)
  • Add validation feedback (re-score after corrections)
  • Create quality trend dashboard (corrections per document type)
  • Polish all UI/UX

Deliverables

  • Inline edit mode for extracted fields
  • Recalculate quality score after edits
  • Corrections audit trail (track what changed and why)
  • Trend dashboard (accuracy by document type, common errors)
  • Complete visual polish (colors, spacing, responsive design)

Team Effort

Engineer: Correction interface, recalculation logic, styling

Architect: Schema validation, audit trail design

Success Criteria:
✓ Corrections fully functional
✓ Audit trail complete
✓ UI polished for stakeholder demo
Day 5 (Friday) — Testing, Documentation & Demo Prep

Goals

  • End-to-end functional testing (upload → extract → quality → correct)
  • Create presentation deck with screenshots
  • Document technical architecture
  • Prepare for stakeholder demo

Deliverables

  • Test results document (functionality checklist)
  • Presentation deck (6-8 slides): Problem → Solution → Demo → Next Steps
  • Architecture diagram (LLM integration, quality engine, persona workflow)
  • Prototype runbook (how to open/use the demo)
  • Stakeholder feedback form (for demo attendees)

Team Effort

Engineer: Testing, documentation

Architect: Presentation, demo walkthrough prep

Success Criteria:
✓ All features verified working
✓ Demo ready for live presentation
✓ Stakeholder feedback captured post-demo

Daily Standups

Communication Cadence

Time: 9:00 AM daily

Attendees: Kyndryl Engineer + Architect + SIVA (if available)

Format:

Blockers & Mitigation

Risk Management
Blocker Mitigation
LLM API rate limits Use cached results; batch processing; fallback to mock data
Sample documents unavailable Use synthetic sample PDFs (generated via LLM)
Azure environment access delays Build locally first; integrate Azure on Day 3-4
Stakeholder availability for feedback Pre-demo review on Day 4; adjust based on feedback on Day 5

Prototype Features Demo (End of Day 5)

Stakeholder Presentation
  1. Upload Workflow: Drag-and-drop policy document
  2. Extraction Results: Show extracted structured fields with confidence scores
  3. Quality Dashboard: Visual indicators (✓ High / ⚠ Low / ✗ Missing)
  4. Claims Adjuster View: Highlight coverage-relevant fields
  5. Underwriter View: Highlight risk assessment fields
  6. Side-by-Side: Original document + extracted data with highlighting
  7. Corrections: Edit misextractions, recalculate quality
  8. Trends: Show common extraction errors by document type

Stakeholder Feedback Questions

  1. Is the extraction workflow intuitive?
  2. Does quality scoring make sense?
  3. Are the claims adjuster and underwriter views useful?
  4. What fields are we missing or getting wrong?
  5. What would make this production-ready?

Sprint Success Criteria (End of Sprint)

Criterion Status
All extraction features working
Quality dashboard functional
Dual personas implemented
Side-by-side comparison working
Human correction workflow complete
Prototype runs without server
Presentation ready for stakeholders
Feedback form prepared
Phase 5 scope refined based on feedback

Post-Sprint (Week 2)

Phase Transition
Monday (3/17): Stakeholder demo + feedback collection
Tuesday-Wednesday (3/18-19): Refine scope based on feedback
Thursday-Friday (3/20-21): Prepare Phase 5 technical plan and resource allocation

Team Capacity

Role Availability Effort
Architect 0.5 FTE (design, review) ~20 hours
Engineer 1.0 FTE (build, test) ~40 hours
QA/Testing 0.2 FTE (verify) ~8 hours
Total ~68 hours

View Related Documents

← Return to Table of Contents for Architecture Diagram, POC Specification, and other project materials.