feat: simplify document insertion

README.md

@@ -10,12 +10,12 @@ RAGLite is a Python package for Retrieval-Augmented Generation (RAG) with SQLite
 2. 🔒 Fully local RAG with [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) as an LLM provider and [SQLite](https://github.com/sqlite/sqlite) as a local database
 3. 🚀 Acceleration with Metal on macOS and with CUDA on Linux and Windows
 4. 📖 PDF to Markdown conversion on top of [pdftext](https://github.com/VikParuchuri/pdftext) and [pypdfium2](https://github.com/pypdfium2-team/pypdfium2)
-5. ✂️ Optimal [level 4 semantic chunking](https://medium.com/@anuragmishra_27746/five-levels-of-chunking-strategies-in-rag-notes-from-gregs-video-7b735895694d)
+5. ✂️ Optimal [level 4 semantic chunking](https://medium.com/@anuragmishra_27746/five-levels-of-chunking-strategies-in-rag-notes-from-gregs-video-7b735895694d) by solving a [binary integer programming problem](https://en.wikipedia.org/wiki/Integer_programming)
 6. 📌 Markdown-based [contextual chunk headings](https://d-star.ai/solving-the-out-of-context-chunk-problem-for-rag)
-7. 🌈 [Multi-vector chunk retrieval](https://python.langchain.com/v0.2/docs/how_to/multi_vector/)
-8. 🌀 Optimal closed-form linear query adapter by solving an [(orthogonal) Procrustes problem](https://en.wikipedia.org/wiki/Orthogonal_Procrustes_problem)
+7. 🌈 Sub-chunk matching with [multi-vector chunk retrieval](https://python.langchain.com/v0.2/docs/how_to/multi_vector/)
+8. 🌀 Optimal [closed-form linear query adapter](src/raglite/_query_adapter.py) by solving an [orthogonal Procrustes problem](https://en.wikipedia.org/wiki/Orthogonal_Procrustes_problem)
 9. 🔍 [Hybrid search](https://plg.uwaterloo.ca/~gvcormac/cormacksigir09-rrf.pdf) that combines [SQLite's BM25 full-text search](https://sqlite.org/fts5.html) with [PyNNDescent's ANN vector search](https://github.com/lmcinnes/pynndescent)
-10. ✍️ Optional support for automatic conversion of any input document to Markdown with [Pandoc](https://github.com/jgm/pandoc)
+10. ✍️ Optional support for conversion of any input document to Markdown with [Pandoc](https://github.com/jgm/pandoc)
 
 ## Installing
 
@@ -49,11 +49,10 @@ my_config = RAGLiteConfig(db_url="sqlite:///raglite.sqlite")
 
 # Index documents:
 from pathlib import Path
-from raglite import insert_document, update_vector_index
+from raglite import insert_document
 
 insert_document(Path("On the Measure of Intelligence.pdf"), config=my_config)
-insert_document(Path("Situational Awareness.pdf"), config=my_config)
-update_vector_index(config=my_config)
+insert_document(Path("Special Relativity.pdf"), config=my_config)
 
 # Search for chunks:
 from raglite import hybrid_search, keyword_search, vector_search
@@ -66,7 +65,7 @@ results_hybrid = hybrid_search(prompt, num_results=5, config=my_config)
 # Answer questions with RAG:
 from raglite import rag
 
-prompt = "What is a 'SkillProgram'?"
+prompt = "What does it mean for two events to be simultaneous?"
 stream = rag(prompt, search=hybrid_search, config=my_config)
 for update in stream:
     print(update, end="")

src/raglite/_index.py

@@ -54,8 +54,10 @@ def _create_chunk_records(
     return chunk_records
 
 
-def insert_document(doc_path: Path, *, config: RAGLiteConfig | None = None) -> None:
-    """Insert a document into the database."""
+def insert_document(
+    doc_path: Path, *, update_index: bool = True, config: RAGLiteConfig | None = None
+) -> None:
+    """Insert a document into the database and update the index."""
     # Use the default config if not provided.
     config = config or RAGLiteConfig()
     # Preprocess the document into chunks.
@@ -96,6 +98,9 @@ def insert_document(doc_path: Path, *, config: RAGLiteConfig | None = None) -> N
                 continue
             session.add(chunk_record)
             session.commit()
+    # Update the vector search chunk index.
+    if update_index:
+        update_vector_index(config)
 
 
 def update_vector_index(config: RAGLiteConfig | None = None) -> None: