AI-RAG-docuquery: my AI-based document research project
AI-RAG-docuquery is an application that combines artificial intelligence and semantic search to help quickly and naturally consult your own…
AI-RAG-docuquery is an application that combines artificial intelligence and semantic search to help quickly and naturally consult your own local documents.
It is based on the Retrieval-Augmented Generation (RAG) paradigm: an approach that combines the retrieval of relevant texts with natural language generation.
GitHub repo: github.com/paoloronco/AI-RAG-docuquery-app
In practice, the user:
- Indexes their files (PDF, Word, Excel, PowerPoint, simple text, Markdown, CSV).
- Asks natural language questions through a simple GUI interface (PyQt6).
- Receives concise and cited answers, with direct links to the original sources.
This means transforming a folder of documents into some sort of personal assistant, capable of answering questions and guiding document consultation without having to open and manually search them.
Why it’s useful
Many professionals and students find themselves managing large amounts of documents: technical manuals, lecture notes, PDF archives. Searching for information in these files is often slow and frustrating.
With AI-RAG-docuquery:
- Time is saved thanks to smarter semantic searches than traditional text-based searches.
- Answers are always linked to verifiable sources, so one doesn’t have to blindly trust the model.
- It can use both local HuggingFace models (offline) and OpenAI-compatible services, making the system flexible to user needs.
- It works in a “No LLM” mode, i.e., without language models, for those who just want the most similar passages found in the documents.
Main features
- FAISS vector search engine for document indexing.
- Multi-format support (PDF, DOCX, PPTX, XLSX, TXT, CSV, MD).
- Intuitive PyQt6 graphical user interface, cross-platform.
- Simplified OpenAI configuration: dedicated popup with API key management and model selection.
- Compatibility with local HuggingFace, which automatically utilizes CPU or GPU.
- Executable build (.exe) via PyInstaller, so the app can be distributed without requiring Python installation.
Skills acquired through the project
The realization of this project allowed me to work on various technical fronts, consolidating and expanding my skills:
- NLP and applied AI
- Understanding the RAG paradigm and its practical use.
- Using Sentence-Transformers for embeddings and HuggingFace models.
- Integration with external model APIs (OpenAI/compatible).
- Information Retrieval
- Managing vector indexes with FAISS.
- Hybrid retrieval techniques (dense + sparse).
- Creating pipelines for content extraction from heterogeneous formats.
- Python software development
- Modular code organization (indexer, retrieve, loaders, llm_clients).
- Dependency management and virtual environments.
- Distribution via PyInstaller with optimizations for heavy libraries like torch and faiss.
- GUI development
- Creating graphical interfaces with PyQt6.
- Integrating advanced features such as clickable links that open local files.
- Persisting user configurations through JSON files.
- Software engineering & DevOps
- Version control with Git and distribution on GitHub.
- Package management, cross-platform troubleshooting, and Windows optimization.
- Full documentation with README and roadmap.
Conclusion
AI-RAG-docuquery is a project that combines academic research with practicality: an app to simplify the consultation of personal documents, but also an advanced exercise in software engineering and AI integration.