studyengine 0.1.0

Retrieval-augmented study engine: document parsing, structure-aware chunking, hybrid retrieval, LLM backends, and generated study tools.

studyengine

Retrieval-augmented study engine. Document parsing, structure-aware chunking, hybrid retrieval, LLM backends, and generated study tools (summaries, quizzes, flashcards, close reading).

Install

pip install studyengine

Python 3.11+.

Usage

Point storage at an app-owned directory once, then ingest and retrieve.

import studyengine

studyengine.configure("./storage")  # creates chroma/, embed_cache.sqlite, covers/, markdown/

from studyengine.parser import parse
from studyengine.chunker import chunk_document
from studyengine import vectorstore
from studyengine.retriever import retrieve

doc = parse("paper.pdf")
chunks = chunk_document(doc.document)
vectorstore.add_chunks(doc_id="paper", chunks=chunks)

hits = retrieve("what is the main claim?", doc_ids=["paper"])

Modules

Module	Purpose
parser	PDF/DOCX/text parsing via docling
sections	Heading-aware section detection
chunker	Structure-aware, token-budgeted chunking
embedder	Sentence-transformer embeddings with on-disk cache
vectorstore	Chroma-backed storage and hybrid query
retriever	Dense + BM25 retrieval fused with RRF
composer	Prompt assembly within a context budget
summarizer	Per-section summaries
quiz	Question generation and grading
close_reading	Scoped chat, comprehension, and "go deeper" streams
llm	Anthropic and Ollama backends behind a common interface

Configuration

configure(root) sets the storage layout. Backends and model choices read from environment variables (LLM_BACKEND, ANTHROPIC_MODEL, EMBED_MODEL, OLLAMA_BASE_URL, and others in studyengine.config).