feat(pattern, rig): dictionary & llm-driven identification#26
Merged
feat(pattern, rig): dictionary & llm-driven identification#26
Conversation
…gories Ensure all dependency versions specify major.minor, add tracing-subscriber to workspace dependencies, sort members and internal crates alphabetically, and fix dependency category groupings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Rename ServerConfig to Cli as top-level parser, extract ServerConfig into config/server.rs for network binding. Split server/ into listen.rs and shutdown.rs, add shutdown timeout with structured tracing, move init_tracing to Cli, and use anyhow::Result for error propagation. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
martsokha
added
docs
…crates Reorganize nvisy-identify from modality-based layout (text/, image/) to detection-method-based layout (pattern/, ner/, llm/, vision/, audio/, fusion/) so the module structure mirrors identification strategies. - Create nvisy-ocr crate: OcrBackend trait, config, parsing, PythonBridge - Create nvisy-asr crate: TranscribeBackend trait, config, parsing, PythonBridge - Add LlmBackend trait and parse_llm_entities to nvisy-rig - Update nvisy-augment to import from nvisy-ocr/nvisy-asr - Add LLM contextual detection layer (llm/detection.rs, llm/prompt.rs) - Add OCR detection layer (vision/ocr.rs) - Add audio transcript+NER composite layer (audio/transcript.rs) - Add ensemble fusion with MaxConfidence/WeightedAverage/NoisyOr strategies - Remove stale nvisy-object workspace references - Sort workspace members, deps, Dockerfile crate lists, and changelog Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…onfidence - Narrow nvisy-pattern root exports to only externally-used types (PatternEngine, PatternEngineBuilder, PatternMatch, DetectionSource, ContextRule); move AllowList/DenyList/PatternEngineError/default_engine behind `pub mod engine` for opt-in access - Add `column_confidence` to DictionaryPattern so CSV dictionary columns can have different confidence scores (e.g. full name vs short code) - Track source column index in CsvDictionary via new Dictionary::columns() - Apply column-specific confidence in PatternEngine::scan_dict - Update currencies/cryptocurrencies/languages patterns with per-column confidence (full names 0.85, codes 0.55/0.45) - Remove API Status link from root README Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move confidence from a top-level JSON field into the match source objects so each source type owns its own scoring: - RegexPattern gains a `confidence: f64` field (default 1.0) - DictionaryPattern.confidence accepts a number (uniform) or array (per-column) via DictionaryConfidence enum - Remove Pattern::confidence() from the trait — confidence is now read directly from the match source during engine compilation - Remove top-level `confidence` from all 27 pattern JSON definitions - Rename `column_confidence` to `confidence` in dictionary patterns Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Absorb small utility modules (error, retry, metrics, compact) into backend/ and rename structured/ to agent/, reducing module sprawl while keeping all public re-exports intact. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…r wrapper Introduce layered agent architecture: - BaseAgent<M> with builder handling rig-core's typestate for tools - NerAgent<M> replacing StructuredAgent with NER-specific prompts - OcrProvider/CvProvider traits in their respective agent modules - ResponseParser as Cow<str> wrapper with extract_text constructor - Stub modules for ocr, cv, and redactor agents Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…edactionMethod Implement the three remaining stub agents in nvisy-rig: - OcrAgent: VLM agent with OcrProvider-backed tool, extracts text from images and detects entities via OcrPromptBuilder - CvAgent: VLM agent with CvProvider-backed tool, detects faces/plates/ signatures via CvPromptBuilder - RedactorAgent: pure LLM agent that recommends TextRedactionMethod for each detected entity via RedactorPromptBuilder Ontology changes (nvisy-ontology): - Rename spec/ to specification/ - Split mod.rs into input.rs (*RedactionInput enums + RedactorInput) and method.rs (TextRedactionMethod, ImageRedactionMethod, AudioRedactionMethod, RedactionMethod) Rig structural changes (nvisy-rig): - Rename agent dirs: ner→recognize, ocr→extract, cv→detect - Flatten agent/mod.rs re-exports (no pub submodules) - Add PromptBuilder structs for all agents (OcrPromptBuilder, CvPromptBuilder, RedactorPromptBuilder) - Add base64 and thiserror dependencies - Improve docs and tracing across all agents Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- BaseAgent.prompt_text() now uses agent.completion() instead of
building raw requests from the model, so preamble/tools/config
are preserved
- Remove model: Arc<M> from BaseAgent (agent owns it)
- Remove system: Option<&str> param from prompt methods (preamble is
on the agent)
- Replace BaseAgentConfig field with context_window: Option<ContextWindow>
since temperature/max_tokens are baked into the rig Agent at build time
- Split base.rs into base/{agent,builder,context}.rs
- Rename redactor/ → redact/ to match action-verb convention
- OcrProvider returns Vec<OcrTextRegion> with bbox support
- Add fn new() constructors to OcrRigTool and CvRigTool
- Add from_prompt error mapper for rig::PromptError
- Export OcrTextRegion from lib.rs and prelude
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add BaseAgent.id (UUIDv7) for observability; expose id() on all specialized agents and include agent_id in tracing spans - Make RetryPolicy generic over any Req: Clone + Res instead of hardcoding DetectionRequest/DetectionResponse - Use : instead of — as doc separator - Use 0.0..=1.0 range notation in confidence docs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…efactor Fix UTF-8 panics in split_to_fit/truncate_to_fit by snapping byte positions to char boundaries. Rewrite prompt_structured to use completion()+output_schema so usage is always recorded. Refactor RigBackend into generic ServiceBackend<S> wrapping any inner Tower service with usage tracking and tracing. Export BaseAgentConfig and ContextWindow for external consumers. Add Clone+PartialEq to all public output types. Restrict from_completion/from_prompt to pub(crate). Deduplicate ALL_TYPES_HINT. Remove dead parse_json_array. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… remove RedactorAgent - Add LLM-based compact() on ContextWindow and prompt_compact() on BaseAgent for summarizing text that exceeds the token budget - Delete nvisy-ocr crate; move OcrBackend, OcrConfig, parse_ocr_entities, and PythonBridge impl into nvisy-rig/src/paddle module - Update nvisy-identify and nvisy-augment to import from nvisy_rig::paddle - Remove RedactorAgent, keeping NerAgent, OcrAgent, and CvAgent - Clean up workspace Cargo.toml, Dockerfile, and all re-exports Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…paddle crate Move OCR backend code out of nvisy-rig/src/paddle/ into a new nvisy-paddle crate so nvisy-rig no longer depends on nvisy-python. Consumers (nvisy-identify, nvisy-augment) now import from nvisy_paddle. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… src/error.rs Add a proper Error enum that implements From<CompletionError>, From<PromptError>, and Into<nvisy_core::Error>. Delete the old backend/error.rs helper functions and update all call sites. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…h plain connection params Replace all CompletionModel generics with a Provider enum holding connection parameters (api_key, base_url). Client construction is deferred to build time via ProviderClient. Agent and backend constructors now return Result to propagate client errors instead of panicking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace Tower service layer with reqwest-middleware + reqwest-retry for transparent HTTP-level retries. Delete ServiceBackend, RigBackend, RetryPolicy, and dispatch_model! macro. Replace tower::Service bound in nvisy-identify with LlmBackend async trait. Rename agent submodules: detect→cv, extract→ocr, recognize→ner. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…aw* types, move compact to BaseAgent Extract max_retries from provider structs into standalone RetryConfig. Replace HttpClient type alias with ClientWithMiddleware directly. Rename entity types: RawEntity→NerEntity, RawCvEntity→CvEntity, RawOcrEntity→OcrEntity. Move compact logic from ContextWindow to BaseAgent::prompt_compact where it belongs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… RetryConfig into BaseAgentConfig Fold client construction directly into Agents::build(), eliminating the ProviderClient intermediary. Move model_name from a separate parameter into Provider variants so each provider carries its full identity. Merge max_retries into BaseAgentConfig, removing the standalone RetryConfig struct. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ivial tests Move agent/base/* files (BaseAgent, BaseAgentBuilder, BaseAgentConfig, ContextWindow, Provider) into backend/ so the agent infrastructure lives alongside usage tracking and detection types. Make the agent module private (was pub(crate)) and re-export public types through backend/. Improve module and type documentation across the crate. Remove 9 trivial tests that only verified arithmetic or getters (23 → 14 tests). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… delete EntityParser/vision/ontology - Add reqwest-tracing middleware and 120s timeout to HTTP client - Move base agent from backend/agent/ to agent/base/ module - Delete EntityParser from nvisy-rig, inline logic in nvisy-identify - Delete vision/ and ontology/ modules from nvisy-identify - Make all internal modules private, re-export from parent mod.rs - Remove nvisy-paddle dependency from nvisy-identify Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ema for tool args - Fix nvisy_ontology::spec → nvisy_ontology::specification in engine test - Replace hand-written json!() tool schemas with schemars::schema_for!() - Add Debug, Clone, JsonSchema derives to CvToolArgs and OcrToolArgs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…y-server - Add missing features = [] to reqwest-middleware, reqwest-retry, reqwest-tracing in workspace Cargo.toml - Remove pub use re-exports (routes, ServiceState) from nvisy-server - Update nvisy-cli to use full module paths Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ed offset resolution, and KnownNerEntity accumulation Move preamble into BaseAgentConfig so specialized agents set it via config. Redesign NerEntity with entity_id for coreference, optional category/entity_type/confidence, context snippet for deterministic offset resolution, and LLM-produced description. Add KnownNerEntity for lightweight cross-chunk context, NerContext with merge/set_text for accumulating surface forms and descriptions across calls, and ResolvedOffsets with type-safe resolve_offsets tied to the source NerContext. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
martsokha
changed the title
martsokha
changed the title
… adapters Delete the old detection modules that duplicated logic now provided by nvisy-rig and nvisy-pattern. Replace them with thin adapter structs in a new method/ module: NerMethod (wraps NerAgent), CvMethod (wraps CvAgent), and PatternDetection (migrated as-is). Remove nvisy-python and bytes deps that were only needed by the deleted code. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.