Skip to content

CHANGELOG.md

Latest commit

 

History

History
116 lines (97 loc) · 4.28 KB

CHANGELOG.md

File metadata and controls

116 lines (97 loc) · 4.28 KB

Changelog

All notable changes to QuantumRAG will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.1.0] - 2026-03-25

Added

Core Engine

  • Engine class as single entry point for all RAG functionality
  • Sync and async interfaces for all operations
  • Adaptive complexity routing (Simple/Medium/Complex query paths)
  • Configurable via YAML, environment variables, and code arguments

Triple Index Architecture

  • Original Embedding index (cosine similarity via LanceDB)
  • HyPE (Hypothetical Prompt Embeddings) index for question-to-question matching
  • Contextual BM25 index (tantivy-py) with Korean morphology support
  • Reciprocal Rank Fusion (RRF) with configurable weights

Document Ingestion

  • Multi-format parsing: PDF, DOCX, PPTX, XLSX, HTML, Markdown, CSV, HWP/HWPX, plain text
  • Three chunking strategies: fixed-size, semantic, structural (auto-detection)
  • Contextual prefix generation for enhanced BM25 indexing
  • Quality scoring (0.0-1.0) for parsed documents
  • Incremental indexing (hash-based change detection)

Retrieval Pipeline

  • Triple Index fusion search with RRF scoring
  • FlashRank reranking (CPU, free)
  • Extractive context compression
  • Metadata-based filtering
  • Full processing trace for debugging

Generation

  • Source-grounded generation with inline citations [1], [2]
  • Korean/English bilingual prompt system
  • Confidence assessment (strongly_supported, partially_supported, insufficient_evidence)
  • "I don't know" responses when evidence is insufficient
  • Streaming support (SSE)

LLM Providers

  • OpenAI (GPT-4.1 family, embeddings)
  • Anthropic (Claude family)
  • Ollama (local models, no API key needed)
  • Automatic retry with exponential backoff
  • Cost tracking and usage monitoring

Korean Language Support

  • HWP/HWPX document parsing
  • Kiwi morphology analyzer integration
  • EUC-KR/CP949 encoding detection and conversion
  • Korean-aware query routing (substring matching for agglutinative morphology)

HTTP API (FastAPI)

  • POST /v1/ingest, /v1/query, /v1/query/stream
  • GET /v1/documents, /v1/status
  • DELETE /v1/documents/:id
  • POST /v1/evaluate, /v1/feedback
  • API key authentication
  • Rate limiting
  • CORS configuration

Evaluation System

  • Synthetic QA pair generation (template-based, zero cost)
  • Retrieval Recall@K metric
  • Faithfulness metric (sentence-level)
  • Answer Relevancy metric
  • Latency percentiles (P50, P95, P99)
  • Weakness analysis with improvement suggestions

CLI (Typer)

  • quantumrag init - Initialize project
  • quantumrag ingest <path> - Ingest documents
  • quantumrag query "<question>" - Query with citations
  • quantumrag status - Show index status
  • quantumrag serve - Start HTTP API server
  • Rich progress bars and formatted output

Advanced Features

  • Plugin system with hook-based extensibility
  • Semantic cache with TTL and similarity matching
  • Batch query processing with concurrency control
  • Query tracing and cost tracking (observability)

Connectors

  • Local file system
  • URL/web page
  • AWS S3

[0.2.0] - 2026-03-29

Added

  • Post-generation correction pipeline (retrieval retry → self-correct → fact verify → completeness)
  • Fact verifier: rule-based hallucination detection via fact cross-checking (zero LLM cost)
  • Completeness checker: multi-part answer verification with targeted re-retrieval
  • Query classifier for adaptive routing
  • Autotune framework for parameter optimization
  • Chunk coherence scoring and input denoising
  • Chroma and FAISS vector store backends
  • Map-Reduce RAG for aggregation queries
  • QA dataset framework with 4 datasets (105 questions), auto-graduation
  • Combined QA runner for retrieval precision testing at scale
  • Scenario test suite (v1-v4, 176 test cases)
  • Scale test framework
  • Makefile with tiered test strategy (quick/smoke/check/scenario)
  • GitHub Actions CI: mypy type checking, security tests, coverage threshold

Removed

  • Multi-tenant support (isolated storage) — removed, not needed for current use cases
  • Query decomposition module (decomposer.py) — replaced by query expander
  • Speculative RAG (speculative.py) — replaced by post-generation correction pipeline
  • Evidence extractor — functionality merged into fact extractor