Stabilize PDF fixture conversion

.gitignore

@@ -6,3 +6,7 @@ test.pdf
 test_output.md
 images/
 examples/
+pdfs/*
+!pdfs/.gitkeep
+output/*
+!output/.gitkeep

Cargo.toml

@@ -31,3 +31,6 @@ scraper = "0.21"
 dirs = "5"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
+
+[dev-dependencies]
+regex = "1"

README.md

@@ -8,11 +8,11 @@ CLI utility in Rust for converting various sources to Markdown. Supports PDF fil
 
 Download the latest release from the [Releases page](https://github.com/Ivlad003/any2md/releases/latest).
 
-| Platform | File |
-|----------|------|
+| Platform                      | File                                   |
+| ----------------------------- | -------------------------------------- |
 | macOS (Apple Silicon & Intel) | `any2md-vX.Y.Z-macos-universal.tar.gz` |
-| Linux x86_64 | `any2md-vX.Y.Z-linux-x86_64.tar.gz` |
-| Windows x86_64 | `any2md-vX.Y.Z-windows-x86_64.zip` |
+| Linux x86_64                  | `any2md-vX.Y.Z-linux-x86_64.tar.gz`    |
+| Windows x86_64                | `any2md-vX.Y.Z-windows-x86_64.zip`     |
 
 ```bash
 # macOS / Linux
@@ -27,26 +27,33 @@ Expand-Archive any2md-*.zip -DestinationPath .
 
 ### Build from Source
 
+Use this when developing locally or when you want to install the CLI from a checked-out copy of the repository.
+
 #### Prerequisites
 
-| Feature | Requirement |
-|---------|-------------|
-| PDF | None (built-in) |
-| Image OCR (local) | [Tesseract](https://github.com/tesseract-ocr/tesseract) installed |
-| Image OCR (cloud) | `OPENAI_API_KEY` environment variable |
-| Audio (local) | Auto-downloads Whisper model on first use. Requires `cmake` at build time. |
-| Audio (cloud) | `OPENAI_API_KEY` environment variable |
-| Website | None (built-in) |
+| Feature           | Requirement                                                                |
+| ----------------- | -------------------------------------------------------------------------- |
+| PDF               | None (built-in)                                                            |
+| Image OCR (local) | [Tesseract](https://github.com/tesseract-ocr/tesseract) installed          |
+| Image OCR (cloud) | `OPENAI_API_KEY` environment variable                                      |
+| Audio (local)     | Auto-downloads Whisper model on first use. Requires `cmake` at build time. |
+| Audio (cloud)     | `OPENAI_API_KEY` environment variable                                      |
+| Website           | None (built-in)                                                            |
 
 ```bash
 # macOS
 brew install tesseract cmake
 
 # Ubuntu/Debian
 sudo apt install tesseract-ocr cmake
+```
 
-# Build
-cargo build --release
+#### Install Locally
+
+From the repository root:
+
+```bash
+cargo install --path .
 ```
 
 The binary will be at `target/release/any2md`.
@@ -111,8 +118,8 @@ any2md document.pdf --debug
 Regular paragraph text with **bold** and *italic* formatting.
 
 | Column 1 | Column 2 | Column 3 |
-| --- | --- | --- |
-| Data | Data | Data |
+| -------- | -------- | -------- |
+| Data     | Data     | Data     |
 
 - List item one
 - List item two
@@ -262,12 +269,12 @@ Good point. Let me pull up the metrics from last month.
 
 ## Supported Formats
 
-| Format | Engine | Notes |
-|--------|--------|-------|
-| PDF | Built-in (`lopdf`) | 4-phase pipeline: extract, detect tables, classify, assemble |
-| Website | `reqwest` + `scraper` | Reader-mode extraction, SSRF protection |
-| Image OCR | Tesseract CLI / OpenAI Vision | Local or cloud via `--engine` flag |
-| Audio | Whisper.cpp / OpenAI Whisper API | Local or cloud, file or live mic |
+| Format    | Engine                           | Notes                                                        |
+| --------- | -------------------------------- | ------------------------------------------------------------ |
+| PDF       | Built-in (`lopdf`)               | 4-phase pipeline: extract, detect tables, classify, assemble |
+| Website   | `reqwest` + `scraper`            | Reader-mode extraction, SSRF protection                      |
+| Image OCR | Tesseract CLI / OpenAI Vision    | Local or cloud via `--engine` flag                           |
+| Audio     | Whisper.cpp / OpenAI Whisper API | Local or cloud, file or live mic                             |
 
 ## Architecture
 
@@ -355,16 +362,16 @@ This triggers the release workflow which builds binaries for all platforms and c
 
 ## Dependencies
 
-| Crate | Purpose |
-|-------|---------|
-| `lopdf` | PDF parsing |
-| `whisper-rs` | Local speech-to-text (whisper.cpp bindings) |
-| `cpal` | Cross-platform audio capture |
-| `symphonia` | Audio format decoding (MP3, OGG, FLAC, WAV, AAC) |
-| `reqwest` | HTTP client (web fetch, cloud APIs) |
-| `scraper` | HTML DOM parsing |
-| `clap` | CLI argument parsing |
-| `tracing` | Structured logging |
-| `serde_json` | JSON parsing for cloud API responses |
-| `base64` | Base64 encoding for inline images and cloud OCR |
-| `dirs` | Home directory resolution for model storage |
+| Crate        | Purpose                                          |
+| ------------ | ------------------------------------------------ |
+| `lopdf`      | PDF parsing                                      |
+| `whisper-rs` | Local speech-to-text (whisper.cpp bindings)      |
+| `cpal`       | Cross-platform audio capture                     |
+| `symphonia`  | Audio format decoding (MP3, OGG, FLAC, WAV, AAC) |
+| `reqwest`    | HTTP client (web fetch, cloud APIs)              |
+| `scraper`    | HTML DOM parsing                                 |
+| `clap`       | CLI argument parsing                             |
+| `tracing`    | Structured logging                               |
+| `serde_json` | JSON parsing for cloud API responses             |
+| `base64`     | Base64 encoding for inline images and cloud OCR  |
+| `dirs`       | Home directory resolution for model storage      |

src/converter/audio/mod.rs

@@ -650,6 +650,7 @@ fn build_document(title: Option<String>, sections: &[SpeakerSection]) -> Documen
             title,
             author: None,
             date: None,
+            date_label: None,
         },
         pages: vec![Page { elements }],
     }

src/converter/image_ocr/mod.rs

@@ -58,6 +58,7 @@ impl ImageOcrConverter {
                 title,
                 author: None,
                 date: None,
+                date_label: None,
             },
             pages: vec![Page { elements }],
         };

src/converter/pdf/assembler.rs

@@ -40,6 +40,9 @@ fn is_header_footer_noise(text: &str) -> bool {
     if t.is_empty() {
         return true;
     }
+    if Assembler::is_recurring_page_chrome(t) {
+        return true;
+    }
     // Standalone "OneNote"
     if t == "OneNote" {
         return true;
@@ -84,6 +87,16 @@ fn starts_lowercase(text: &str) -> bool {
 }
 
 impl Assembler {
+    pub fn is_recurring_page_chrome(text: &str) -> bool {
+        let normalized = text
+            .trim()
+            .replace('fi', "fi")
+            .replace('é', "e")
+            .replace('è', "e")
+            .to_lowercase();
+        normalized == "picto erp - specifications fonctionnelles"
+    }
+
     pub fn assemble(
         classified_pages: Vec<Vec<ClassifiedElement>>,
         metadata: Metadata,
@@ -121,13 +134,13 @@ impl Assembler {
 
         for page in pages {
             let should_merge = if let Some(prev_page) = result.last() {
-                matches!(
-                    (prev_page.elements.last(), page.elements.first()),
+                match (prev_page.elements.last(), page.elements.first()) {
                     (
                         Some(Element::Table { headers: h1, .. }),
                         Some(Element::Table { headers: h2, .. }),
-                    ) if h1.len() == h2.len()
-                )
+                    ) => h1.len() == h2.len() || (h1.len() >= 8 && h2.len() >= 8),
+                    _ => false,
+                }
             } else {
                 false
             };
@@ -143,10 +156,16 @@ impl Assembler {
                 {
                     // Merge into the previous page's last table
                     if let Some(prev_page) = result.last_mut() {
-                        if let Some(Element::Table { rows, .. }) = prev_page.elements.last_mut() {
+                        if let Some(Element::Table { headers, rows }) =
+                            prev_page.elements.last_mut()
+                        {
                             // The "header" of the continuation is really a data row
-                            rows.push(next_headers);
-                            rows.append(&mut next_rows);
+                            rows.push(Self::fit_row_to_columns(next_headers, headers.len()));
+                            rows.extend(
+                                next_rows
+                                    .drain(..)
+                                    .map(|row| Self::fit_row_to_columns(row, headers.len())),
+                            );
                         }
                     }
                 }
@@ -164,6 +183,25 @@ impl Assembler {
         result
     }
 
+    fn fit_row_to_columns(mut row: Vec<String>, columns: usize) -> Vec<String> {
+        if row.len() > columns {
+            let extras = row.split_off(columns);
+            if let Some(last) = row.last_mut() {
+                for extra in extras {
+                    if !extra.trim().is_empty() {
+                        if !last.trim().is_empty() {
+                            last.push(' ');
+                        }
+                        last.push_str(extra.trim());
+                    }
+                }
+            }
+        } else {
+            row.resize(columns, String::new());
+        }
+        row
+    }
+
     fn assemble_page(elems: Vec<ClassifiedElement>, metrics: &PageMetrics) -> Page {
         let mut elements = Vec::new();
         let mut i = 0;
@@ -323,9 +361,10 @@ impl Assembler {
                         let mut result_block = block.clone();
                         result_block.has_bold = current_bold;
                         result_block.has_italic = current_italic;
-                        elements.push(Element::Paragraph {
-                            text: Self::rich_text_from_block(&para_text, &result_block),
-                        });
+                        for paragraph in Self::split_key_value_paragraph(&para_text, &result_block)
+                        {
+                            elements.push(paragraph);
+                        }
                     }
                 },
             }
@@ -395,6 +434,38 @@ impl Assembler {
             }],
         }
     }
+
+    fn split_key_value_paragraph(text: &str, block: &RawTextBlock) -> Vec<Element> {
+        const LABELS: [&str; 5] = ["Status:", "Name FR:", "Name EN:", "Sources:", "Figma:"];
+
+        let mut positions: Vec<(usize, &str)> = LABELS
+            .iter()
+            .filter_map(|label| text.find(label).map(|pos| (pos, *label)))
+            .collect();
+        positions.sort_by_key(|(pos, _)| *pos);
+
+        if positions.len() < 2 || positions.first().map(|(pos, _)| *pos) != Some(0) {
+            return vec![Element::Paragraph {
+                text: Self::rich_text_from_block(text, block),
+            }];
+        }
+
+        let mut elements = Vec::with_capacity(positions.len());
+        for (idx, (start, _)) in positions.iter().enumerate() {
+            let end = positions
+                .get(idx + 1)
+                .map(|(next_start, _)| *next_start)
+                .unwrap_or_else(|| text.len());
+            let segment = text[*start..end].trim();
+            if !segment.is_empty() {
+                elements.push(Element::Paragraph {
+                    text: Self::rich_text_from_block(segment, block),
+                });
+            }
+        }
+
+        elements
+    }
 }
 
 #[cfg(test)]
@@ -436,6 +507,7 @@ mod tests {
             title: None,
             author: None,
             date: None,
+            date_label: None,
         }
     }
 

src/converter/pdf/mod.rs

@@ -56,22 +56,30 @@ impl Converter for PdfConverter {
                 // with a Y position greater than the table's Y position
                 let insert_pos = classified
                     .iter()
-                    .position(|el| match el {
-                        ClassifiedElement::Text(b, _) => b.y > table.y_position,
-                        _ => false,
-                    })
+                    .position(
+                        |el| matches!(el, ClassifiedElement::Text(b, _) if b.y > table.y_position),
+                    )
                     .unwrap_or(classified.len());
+                let insert_pos = classified
+                    .iter()
+                    .rposition(|el| matches!(el, ClassifiedElement::Image(_)))
+                    .map(|pos| pos + 1)
+                    .unwrap_or(insert_pos);
                 classified.insert(insert_pos, ClassifiedElement::PreBuilt(table.element));
             }
 
             all_classified.push(classified);
         }
 
         debug!("Phase 3: Building metadata");
+        let title = pdf_meta
+            .title
+            .filter(|title| !Assembler::is_recurring_page_chrome(title));
         let metadata = Metadata {
-            title: pdf_meta.title,
+            title,
             author: pdf_meta.author,
             date: pdf_meta.date,
+            date_label: Some("Created".to_string()),
         };
 
         debug!("Phase 4: Assembling document");