edgequake-pdf2md 0.4.6

Convert PDF documents to Markdown using Vision Language Models — CLI and library
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29

//! Pipeline stages for PDF-to-Markdown conversion.
//!
//! Each submodule implements exactly one transformation step.
//! Keeping stages separate makes each independently testable and lets us
//! swap implementations (e.g. switch rendering backend) without touching
//! other stages.
//!
//! ## Data Flow
//!
//! ```text
//! input ──▶ render ──▶ encode ──▶ llm ──▶ postprocess
//! (URL/path)  (pdfium)  (base64)  (VLM)   (cleanup)
//! ```
//!
//! 1. [`input`]  — canonicalise the user-supplied path or URL to a local file
//! 2. [`render`] — rasterise selected pages; runs in `spawn_blocking` because
//!    pdfium is not async-safe
//! 3. [`encode`] — PNG-encode and base64-wrap each `DynamicImage` for the
//!    multimodal API request body
//! 4. [`llm`]    — drive the VLM call with retry/backoff; the only stage with
//!    network I/O
//! 5. [`postprocess`] — deterministic text-cleanup rules to fix VLM quirks
//!    (markdown fences, hallucinated images, broken tables, etc.)

pub mod encode;
pub mod input;
pub mod llm;
pub mod postprocess;
pub mod render;