| APP_CSS = """ |
| <style> |
| .block-container { |
| max-width: 1080px; |
| padding-top: 2rem; |
| padding-bottom: 4rem; |
| } |
| .doc-hero { |
| padding: 1.45rem 1.55rem; |
| border: 1px solid rgba(120, 120, 120, 0.25); |
| border-radius: 18px; |
| margin-bottom: 1rem; |
| } |
| .doc-eyebrow { |
| font-size: 0.78rem; |
| font-weight: 700; |
| letter-spacing: 0.08em; |
| text-transform: uppercase; |
| opacity: 0.72; |
| margin-bottom: 0.35rem; |
| } |
| .doc-hero h1 { |
| font-size: 2.15rem; |
| line-height: 1.15; |
| margin: 0 0 0.65rem 0; |
| } |
| .doc-hero p { |
| font-size: 1.05rem; |
| line-height: 1.6; |
| margin: 0; |
| opacity: 0.9; |
| } |
| .principle-card, .boundary-card { |
| padding: 1rem 1.1rem; |
| border: 1px solid rgba(120, 120, 120, 0.22); |
| border-radius: 14px; |
| min-height: 100%; |
| } |
| .principle-card strong, .boundary-card strong { |
| display: block; |
| margin-bottom: 0.3rem; |
| } |
| .pipeline-step { |
| padding: 0.85rem 1rem; |
| margin: 0.45rem 0; |
| border-left: 4px solid rgba(100, 120, 220, 0.75); |
| background: rgba(120, 120, 120, 0.06); |
| border-radius: 0 12px 12px 0; |
| } |
| .summary-callout { |
| padding: 1rem 1.1rem; |
| border: 1px solid rgba(120, 120, 120, 0.22); |
| border-radius: 14px; |
| margin-bottom: 0.75rem; |
| } |
| .lineage-note { |
| padding: 0.9rem 1rem; |
| border-radius: 12px; |
| background: rgba(80, 130, 210, 0.08); |
| border: 1px solid rgba(80, 130, 210, 0.22); |
| } |
| </style> |
| """ |
|
|
|
|
| HERO_HTML = """ |
| <div class="doc-hero"> |
| <div class="doc-eyebrow">Agent 5 · Document Intelligence</div> |
| <h1>What did this document say — and can every extracted item be traced back to the source?</h1> |
| <p> |
| Long documents become useful only when people can turn them into structured work products without losing |
| the connection to the original text. This agent parses a document, creates page-aware chunks, extracts |
| lightweight structured intelligence, preserves source lineage, supports search, and exports an auditable record. |
| </p> |
| </div> |
| """ |
|
|
|
|
| WHY_IT_MATTERS_HTML = """ |
| <div class="principle-card"> |
| <strong>The business problem</strong> |
| Analysts often re-read documents, copy facts into spreadsheets, and lose track of which page supported a conclusion. |
| </div> |
| """ |
|
|
|
|
| CONTROL_BOUNDARY_HTML = """ |
| <div class="boundary-card"> |
| <strong>What this agent does</strong> |
| Converts selectable PDF/TXT text into structured, searchable, source-linked outputs using transparent local rules. |
| </div> |
| <div style="height: 0.65rem"></div> |
| <div class="boundary-card"> |
| <strong>What this agent does not claim</strong> |
| It does not prove that the document is true, perform OCR on scanned pages, or replace expert review of consequential documents. |
| </div> |
| """ |
|
|
|
|
| PIPELINE_HTML = """ |
| <div class="pipeline-step"><strong>1 · Parse</strong><br>Read selectable PDF or plain-text content and record extraction notes.</div> |
| <div class="pipeline-step"><strong>2 · Structure</strong><br>Create validated metadata, page-aware chunks, and lightweight summary fields.</div> |
| <div class="pipeline-step"><strong>3 · Preserve lineage</strong><br>Attach extracted facts, dates, entities, risks, and actions to their source page and chunk.</div> |
| <div class="pipeline-step"><strong>4 · Search</strong><br>Search the original document chunks rather than searching only a rewritten summary.</div> |
| <div class="pipeline-step"><strong>5 · Export</strong><br>Produce structured JSON/CSV records that preserve the extraction and source references.</div> |
| """ |
|
|
|
|
| LIMITS_MARKDOWN = """ |
| ### Interpretation boundary |
| |
| The structured findings are **extractions from the supplied document**, not independent verification that the document is correct. |
| |
| - Summary generation uses transparent deterministic rules, not a hosted LLM. |
| - Source lineage means an extracted item can be traced back to the page/chunk where it appeared. |
| - Source lineage does **not** mean the source statement is true. |
| - Scanned PDFs without selectable text may require OCR and are intentionally reported as a limitation rather than silently guessed. |
| - Tables, complex layouts, nuanced entity resolution, and multi-document comparison remain production-upgrade opportunities. |
| - High-stakes legal, medical, financial, compliance, or client documents should receive expert review. |
| """ |
|
|
|
|
| ARCHITECTURE_MARKDOWN = """ |
| ### Engineering architecture |
| |
| ```text |
| PDF / TXT |
| ↓ |
| Deterministic text extraction |
| ↓ |
| Page-aware normalized text |
| ↓ |
| Overlapping document chunks |
| ↓ |
| Transparent structured extraction |
| ├─ executive summary |
| ├─ key facts |
| ├─ entities |
| ├─ dates |
| ├─ risks |
| └─ action items |
| ↓ |
| Source-lineage mapping |
| ├─ page number |
| └─ chunk ID |
| ↓ |
| Validated Pydantic report |
| ├─ search |
| ├─ JSON export |
| └─ CSV export |
| ``` |
| |
| **Control principle:** parsing and extraction are deterministic application logic. The system does not ask an LLM to invent structure or provenance. |
| """ |
|
|
