🔍 Flux

A fast inverted-index search engine with native PDF tokenization — phrase, boolean, and ranked queries out of the box.

Features

• Positional inverted index (term → docId → posting list)
• Phrase search, boolean AND/OR, TF-IDF ranked search
• PDF tokenization (MuPDF) with stop-word filtering
• pybind11 Python bindings (flux module)
• Django backend with PDF.js viewer

Types

struct Token    { std::string text; uint32_t position, pageNo; bool isStopWord; };
struct Posting  { uint32_t position, pageNo; };
struct PhraseHit{ uint32_t docId, position, pageNo; };
struct DocScore { uint32_t docId; double score; };

API

Method	Description
addDocument(docId, tokens)	Index a document's tokens
removeDocument(docId)	Remove a document from the index
getPosting(word)	Get postings for a term
addDocPath(docId, path)	Store file path for a document
getDocCount()	Number of indexed documents
phraseSearch(terms)	Match consecutive terms
andSearch(terms)	Docs containing all terms
orSearch(terms)	Docs containing any term
rankedPhraseSearch(terms)	Phrase match with TF-IDF scoring
rankedAndSearch(terms)	Boolean AND with TF-IDF scoring
rankedOrSearch(terms)	Boolean OR with TF-IDF scoring

Build & test

cmake -B build -S .
cmake --build build
ctest --test-dir build --output-on-failure

Requirements

C++17 · CMake ≥ 3.23 · Python ≥ 3.13

Library	Purpose
MuPDF	PDF parsing
pybind11	C++/Python bindings
GoogleTest	C++ unit testing
Django ≥ 6.0	Web backend

Python: django >= 6.0.5, django-environ >= 0.14.0, pybind11 >= 3.0.4

Project layout

CMakeLists.txt
engine/
├── include/    indexer.hpp, tokenizer.hpp, persist.hpp
├── src/        indexer.cpp, tokenizer.cpp, bindings.cpp, persist.cpp
└── tests/      indexer_test.cpp, tokenizer_test.cpp, test.pdf
config/         Django settings
apps/search/    engine.py, models.py, views.py, urls.py
templates/      HTML templates
static/         CSS, PDF.js viewer
media/          Uploaded documents
.github/        CI workflow