Library

Inspiration

Search engines and AI tools can return answers instantly, but building knowledge still depends on sources we deliberately choose, read, annotate, and revisit. I wanted a simple gateway outside the browser: one focused tool built only for research and developing a personal library as my questions change.

## What it does

Academic Library Builder is a native desktop research workspace.

It lets you:

Search for books, papers, and other learning resources.
Optionally improve a search with GPT-5.6-powered research assistance.
Review discovery results without automatically treating them as owned or trusted.
Deliberately save verified PDFs to My Library.
Read PDFs in the native interface.
Create page-linked annotations without modifying the original document.
Keep a durable, searchable personal knowledge collection.

Discovery results remain separate from My Library until the user explicitly saves or imports a resource.

## How we built it

The application uses Python, PySide6, Qt Widgets, and SQLite. Documents are stored as immutable SHA-256 content objects, while annotations and provenance are maintained separately.

Codex helped drive the engineering process: inspecting the codebase, implementing changes, writing regression tests, debugging cross-layer search behavior, and verifying the packaged native application.

GPT-5.6 powers the optional “Improve results” workflow. It helps refine research intent and organize a stronger reading path, but it cannot save documents, decide rights, prove identity or completeness, or bypass acquisition rules. Those decisions remain explicit and deterministic.

GPT-5.6 was also used as an independent reviewer during final verification, helping identify mismatches between implementation claims and release artifacts.

## Challenges we ran into

The hardest part was keeping different kinds of truth separate. A search result is not necessarily a verified document, and finding a resource does not mean the user owns it.

We also had to make annotations durable without altering source PDFs, handle provider failures honestly, preserve a responsive native interface during research tasks, and package the application with reproducible tests and runtime checks.

## Accomplishments that we're proud of

We built a complete native workflow from question to durable knowledge:

Search for a topic or known work.
Improve the research path with GPT-5.6.
Inspect real sources.
Deliberately save relevant material.
Read and annotate it inside the application.
Return to those annotations later.

The final Linux build passed the complete warning-strict test suite and native packaging checks.

## What we learned

AI is most useful here as a research assistant, not an authority. Models can help clarify intent and navigate unfamiliar literature, but ownership, provenance, rights, and document identity need separate, inspectable rules.

We also learned that a focused native tool can make research feel calmer and more deliberate than a collection of browser tabs.

## What's next for Library

Next steps include stronger annotation search, citation export, reading-path visualization, additional carefully governed discovery providers, and native packaging for more desktop platforms.