refactor + feat: code quality + Presidio-aligned shipped pattern set + chunk context propagation

Cargo.toml

@@ -5,6 +5,7 @@ resolver = "3"
 members = [
     "./crates/nvisy-cli",
     "./crates/nvisy-codec",
+    "./crates/nvisy-context",
     "./crates/nvisy-core",
     "./crates/nvisy-engine",
     "./crates/nvisy-fake",
@@ -37,44 +38,22 @@ documentation = "https://docs.rs/nvisy-runtime"
 
 # Internal crates
 nvisy-codec = { path = "./crates/nvisy-codec", version = "0.1.0", default-features = false }
+nvisy-context = { path = "./crates/nvisy-context", version = "0.1.0" }
 nvisy-core = { path = "./crates/nvisy-core", version = "0.1.0" }
 nvisy-engine = { path = "./crates/nvisy-engine", version = "0.1.0" }
 nvisy-fake = { path = "./crates/nvisy-fake", version = "0.1.0" }
 nvisy-llm = { path = "./crates/nvisy-llm", version = "0.1.0" }
-nvisy-stt = { path = "./crates/nvisy-stt", version = "0.1.0" }
-nvisy-toolkit = { path = "./crates/nvisy-toolkit", version = "0.1.0" }
 nvisy-ner = { path = "./crates/nvisy-ner", version = "0.1.0" }
 nvisy-ocr = { path = "./crates/nvisy-ocr", version = "0.1.0" }
 nvisy-pattern = { path = "./crates/nvisy-pattern", version = "0.1.0" }
 nvisy-server = { path = "./crates/nvisy-server", version = "0.1.0" }
+nvisy-stt = { path = "./crates/nvisy-stt", version = "0.1.0" }
+nvisy-toolkit = { path = "./crates/nvisy-toolkit", version = "0.1.0" }
 
-# Inference & AI frameworks
-bentoml = { version = "0.5", default-features = false, features = ["rustls-tls", "tracing"] }
-rig = { version = "0.38", features = [], default-features = false }
-
-# HTTP client and middleware
-reqwest = { version = "0.13", default-features = false, features = ["json", "rustls", "multipart"] }
-reqwest-middleware = { version = "0.5", features = ["json", "multipart"] }
-reqwest-retry = { version = "0.9", features = [] }
-reqwest-tracing = { version = "0.7", features = [] }
-
-# Async runtime and parallelism
-tokio = { version = "1.50", features = [] }
-tokio-util = { version = "0.7", features = [] }
-futures = { version = "0.3", features = [] }
-async-trait = { version = "0.1", features = [] }
-rayon = { version = "1.10", features = [] }
-
-# Observability
-tracing = { version = "0.1", features = ["attributes"] }
-tracing-subscriber = { version = "0.3", features = [] }
-
-# (De)serialization
+# Serialization
 serde = { version = "1.0", features = ["derive"] }
 serde_json = { version = "1.0", features = [] }
-serde_with = { version = "3.18", features = [] }
 schemars = { version = "1.0", features = ["uuid1", "bytes1"] }
-csv = { version = "1.0", features = [] }
 toml = { version = "1.1", features = [] }
 minijinja = { version = "2.5", features = [] }
 
@@ -92,38 +71,48 @@ hipstr = { version = "0.8", features = ["serde"] }
 jiff = { version = "0.2", features = ["serde"] }
 semver = { version = "1.0", features = ["serde"] }
 oxilangtag = { version = "0.1", features = ["serde"] }
+celes = { version = "2.8", features = [] }
 humantime = { version = "2.1", features = [] }
 humantime-serde = { version = "1.1", features = [] }
 type-map = { version = "0.5", features = [] }
 
 # Encoding and hashing
 base64 = { version = "0.22", features = [] }
+hex = { version = "0.4", features = [] }
 sha2 = { version = "0.11", features = [] }
 aes-gcm = { version = "0.10", features = [] }
-hmac = { version = "0.13", features = [] }
-hex = { version = "0.4", features = [] }
 
-# Pattern matching
+# Async runtime and parallelism
+tokio = { version = "1.50", features = [] }
+tokio-util = { version = "0.7", features = [] }
+futures = { version = "0.3", features = [] }
+async-trait = { version = "0.1", features = [] }
+rayon = { version = "1.10", features = [] }
+
+# Observability
+tracing = { version = "0.1", features = ["attributes"] }
+tracing-subscriber = { version = "0.3", features = [] }
+
+# Text processing (pattern matching, language detection, unicode)
 regex = { version = "1.0", features = [] }
 aho-corasick = { version = "1.0", features = [] }
-smallvec = { version = "1.13", features = [] }
-
-# Language detection and text segmentation
 lingua = { version = "1.8", default-features = false, features = ["english"] }
-stop-words = { version = "0.10", features = ["iso"] }
 unicode-segmentation = { version = "1.13", features = [] }
 unicode-normalization = { version = "0.1", features = [] }
 
-# PDF processing (parsing, text extraction, page-to-image rendering)
-lopdf = { version = "0.41", features = [] }
-pdfium-render = { version = "0.9", features = [] }
+# Checksum / encoding
+bs58 = { version = "0.5", features = ["check"] }
+phonenumber = { version = "0.3", default-features = false }
 
-# Document parsing
+# Tabular document parsing
+csv = { version = "1.0", features = [] }
+calamine = { version = "0.35", features = [] }
+
+# Rich-document parsing (HTML, PDF)
 scraper = { version = "0.27", features = [] }
 ego-tree = { version = "0.11", features = [] }
-calamine = { version = "0.35", features = [] }
-zip = { version = "8.4", features = [] }
-quick-xml = { version = "0.40", features = [] }
+lopdf = { version = "0.41", features = [] }
+pdfium-render = { version = "0.9", features = [] }
 
 # Image processing
 image = { version = "0.25", default-features = false, features = ["png", "jpeg", "tiff"] }
@@ -134,27 +123,30 @@ hound = { version = "3.5", features = [] }
 symphonia = { version = "0.6", default-features = false, features = ["wav", "pcm", "mp3"] }
 mp3lame-encoder = { version = "0.2", features = [] }
 
-# CLI
-clap = { version = "4.6", features = [] }
+# AI / LLM frameworks
+bentoml = { version = "0.5", default-features = false, features = ["rustls-tls", "tracing"] }
+rig = { version = "0.38", features = [], default-features = false }
+
+# HTTP client and middleware
+reqwest-middleware = { version = "0.5", features = ["json", "multipart"] }
+reqwest-retry = { version = "0.9", features = [] }
+reqwest-tracing = { version = "0.7", features = [] }
 
-# HTTP server
+# HTTP server and middleware
 axum = { version = "0.8", features = [] }
 aide = { version = "0.16.0-alpha.2", features = [] }
 tower = { version = "0.5", features = [] }
 tower-http = { version = "0.6", features = [] }
 
-# Filesystem traversal
-walkdir = { version = "2.5", features = [] }
+# CLI
+clap = { version = "4.6", features = [] }
 
-# Storage, file detection, and asset embedding
+# Storage and file-type detection
 fjall = { version = "3.1", features = [] }
-include_dir = { version = "0.7", features = [] }
 infer = { version = "0.19", features = [] }
 
 # Utilities
 validator = { version = "0.20", features = ["derive"] }
 rand = { version = "0.10", features = [] }
 tempfile = { version = "3.27", features = [] }
-
-# Fake data generation
 fake = { version = "5.1", features = [] }

crates/nvisy-cli/Cargo.toml

@@ -58,20 +58,16 @@ path = "src/main.rs"
 nvisy-engine = { workspace = true, features = [] }
 nvisy-server = { workspace = true, features = [] }
 
-# CLI
-clap = { workspace = true, features = ["derive", "env"] }
-
-# (De)serialization
+# Serialization
 serde = { workspace = true, features = [] }
 toml = { workspace = true, features = [] }
-humantime = { workspace = true, features = [] }
 humantime-serde = { workspace = true, features = [] }
 
 # Derive macros and error handling
 anyhow = { workspace = true, features = [] }
 
-# HTTP server
-axum = { workspace = true, features = ["tokio"] }
+# Primitive datatypes
+humantime = { workspace = true, features = [] }
 
 # Async runtime and parallelism
 tokio = { workspace = true, features = ["rt-multi-thread", "macros", "signal"] }
@@ -80,5 +76,11 @@ tokio = { workspace = true, features = ["rt-multi-thread", "macros", "signal"] }
 tracing = { workspace = true, features = [] }
 tracing-subscriber = { workspace = true, features = ["env-filter", "json"] }
 
+# HTTP server and middleware
+axum = { workspace = true, features = ["tokio"] }
+
+# CLI
+clap = { workspace = true, features = ["derive", "env"] }
+
 [package.metadata.cargo-machete]
 ignored = ["humantime-serde"]

crates/nvisy-codec/Cargo.toml

@@ -100,27 +100,39 @@ rustdoc-args = ["--cfg", "docsrs"]
 # Internal crates
 nvisy-core = { workspace = true, features = [] }
 
-# Async runtime
-async-trait = { workspace = true, features = [] }
-tokio = { workspace = true, features = ["sync"] }
-
-# (De)serialization
+# Serialization
 serde = { workspace = true, features = [] }
 serde_json = { workspace = true, features = [] }
+schemars = { workspace = true, features = [] }
+
+# Derive macros and error handling
+derive_more = { workspace = true, features = ["as_ref", "deref", "from"] }
 
 # Primitive datatypes
 bytes = { workspace = true, features = [] }
-uuid = { workspace = true, features = [] }
-derive_more = { workspace = true, features = ["as_ref", "deref", "from"] }
+
+# Encoding and hashing
 hex = { workspace = true, features = [] }
-infer = { workspace = true, features = [] }
-schemars = { workspace = true, features = [] }
 sha2 = { workspace = true, features = [] }
 
-# Image processing — pulled in unconditionally because the image
-# handler structs reference `image::DynamicImage` directly. The
-# workspace dep already enables png/jpeg/tiff decoders. `imageproc`
-# powers the per-region gaussian blur in `image::redact`.
+# Async runtime and parallelism
+async-trait = { workspace = true, features = [] }
+tokio = { workspace = true, features = ["sync"] }
+rayon = { workspace = true, optional = true, features = [] }
+
+# Observability
+tracing = { workspace = true, features = [] }
+
+# Tabular document parsing (feature-gated)
+csv = { workspace = true, optional = true, features = [] }
+
+# Rich-document parsing (feature-gated: HTML + PDF)
+scraper = { workspace = true, optional = true, features = [] }
+ego-tree = { workspace = true, optional = true, features = [] }
+lopdf = { workspace = true, optional = true, features = [] }
+pdfium-render = { workspace = true, optional = true, features = [] }
+
+# Image processing
 image = { workspace = true, features = [] }
 imageproc = { workspace = true, features = [] }
 
@@ -129,18 +141,8 @@ hound = { workspace = true, optional = true, features = [] }
 symphonia = { workspace = true, optional = true, features = [] }
 mp3lame-encoder = { workspace = true, optional = true, features = [] }
 
-# PDF processing (feature-gated)
-lopdf = { workspace = true, optional = true, features = [] }
-pdfium-render = { workspace = true, optional = true, features = [] }
-rayon = { workspace = true, optional = true, features = [] }
-
-# Document parsing (feature-gated)
-csv = { workspace = true, optional = true, features = [] }
-scraper = { workspace = true, optional = true, features = [] }
-ego-tree = { workspace = true, optional = true, features = [] }
-
-# Observability
-tracing = { workspace = true, features = [] }
+# Storage and file-type detection
+infer = { workspace = true, features = [] }
 
 [dev-dependencies]
 tokio = { workspace = true, features = ["macros", "rt"] }

crates/nvisy-codec/src/core/handler.rs

@@ -29,13 +29,21 @@ use crate::content::{ContentData, ContentSource};
 ///
 /// `data` is the per-modality wire payload; `location` is the
 /// coordinate the handler will accept in [`Handler::read`] /
-/// [`Handler::redact`] to address the same chunk again.
+/// [`Handler::redact`] to address the same chunk again. `hints`
+/// carries out-of-band context strings the chunk's structural
+/// neighbours surface — CSV/XLSX column headers, JSON object
+/// keys, HTML parent-element text — for downstream context-aware
+/// recognizers; handlers without such metadata leave it empty.
 #[derive(Debug, Clone, PartialEq)]
 pub struct Chunk<M: Modality> {
     /// Coordinate addressing this chunk inside the handler.
     pub location: M::Location,
     /// Wire payload at the chunk's location.
     pub data: M::Data,
+    /// Out-of-band context strings recognizers should treat as
+    /// in-context (column headers, parent element text, …).
+    /// Empty when the handler has no such metadata to surface.
+    pub hints: Vec<String>,
 }
 
 /// Per-modality capability trait every format handler implements.

crates/nvisy-codec/src/core/loader.rs

@@ -22,6 +22,7 @@
 //! [`Format::loader`]: super::Format::loader
 //! [`FormatId`]: super::FormatId
 
+use std::marker::PhantomData;
 use std::sync::Arc;
 
 use nvisy_core::Error;
@@ -100,7 +101,7 @@ where
 {
     Arc::new(LoaderAdapter {
         loader,
-        _phantom: std::marker::PhantomData,
+        _phantom: PhantomData,
     })
 }
 
@@ -109,7 +110,7 @@ where
 /// [`erase`]; not part of the public API.
 struct LoaderAdapter<M: Modality, L: Loader<M>> {
     loader: L,
-    _phantom: std::marker::PhantomData<fn() -> M>,
+    _phantom: PhantomData<fn() -> M>,
 }
 
 #[async_trait::async_trait]

crates/nvisy-codec/src/document/mod.rs

@@ -29,6 +29,9 @@ mod tabular;
 #[cfg(feature = "internal_text")]
 mod text;
 
+use std::any::type_name;
+use std::fmt;
+
 use derive_more::From;
 #[cfg(feature = "internal_audio")]
 use nvisy_core::modality::Audio;
@@ -226,11 +229,11 @@ impl<M: Modality> DocumentHandle<M> {
     }
 }
 
-impl<M: Modality> std::fmt::Debug for DocumentHandle<M> {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+impl<M: Modality> fmt::Debug for DocumentHandle<M> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         f.debug_struct("DocumentHandle")
             .field("format_id", &self.format_id)
-            .field("modality", &std::any::type_name::<M>())
+            .field("modality", &type_name::<M>())
             .finish()
     }
 }

crates/nvisy-codec/src/handler/audio/mp3_codec.rs

@@ -16,12 +16,13 @@
 //! redaction of MP3 streams. Callers wanting bit-perfect preservation
 //! of unredacted regions should not round-trip.
 
-use std::io::Cursor;
+use std::io::{Cursor, ErrorKind as IoErrorKind};
 
 use bytes::Bytes;
 use mp3lame_encoder::{Builder, FlushNoGap, InterleavedPcm, MonoPcm};
 use nvisy_core::Error;
-use symphonia::core::audio::{Audio as _, GenericAudioBufferRef};
+use symphonia::core::audio::conv::{ConvertibleSample, FromSample};
+use symphonia::core::audio::{Audio as _, AudioBuffer, GenericAudioBufferRef};
 use symphonia::core::codecs::audio::AudioDecoderOptions;
 use symphonia::core::errors::Error as SymError;
 use symphonia::core::formats::probe::Hint;
@@ -139,9 +140,7 @@ pub(super) fn decode_to_pcm(bytes: &Bytes) -> Result<DecodedMp3, Error> {
         let packet = match reader.next_packet() {
             Ok(Some(p)) => p,
             Ok(None) => break,
-            Err(SymError::IoError(io_err))
-                if io_err.kind() == std::io::ErrorKind::UnexpectedEof =>
-            {
+            Err(SymError::IoError(io_err)) if io_err.kind() == IoErrorKind::UnexpectedEof => {
                 break;
             }
             Err(e) => {
@@ -198,22 +197,20 @@ fn append_interleaved_f32(
     channels: usize,
     out: &mut Vec<f32>,
 ) {
-    use symphonia::core::audio::conv::ConvertibleSample;
-
     fn extend<S: ConvertibleSample + Copy>(
-        buf: &symphonia::core::audio::AudioBuffer<S>,
+        buf: &AudioBuffer<S>,
         channels: usize,
         out: &mut Vec<f32>,
     ) where
-        f32: symphonia::core::audio::conv::FromSample<S>,
+        f32: FromSample<S>,
     {
         let frames = buf.frames();
         out.reserve(frames * channels);
         for frame in 0..frames {
             for ch in 0..channels {
                 let plane = buf.plane(ch).expect("plane for known channel index");
                 let sample = plane[frame];
-                out.push(<f32 as symphonia::core::audio::conv::FromSample<S>>::from_sample(sample));
+                out.push(<f32 as FromSample<S>>::from_sample(sample));
             }
         }
     }