Inspiration
What if your library could become a place to explore, rather than merely a place to search?
Over time, our digital libraries become repositories of unfinished conversations: books we once studied, research we set aside, ideas scattered across PDFs, notes, and documents.
We preserve knowledge, yet often lose sight of the connections between its fragments.
SIBILA was born from a simple conviction: knowledge should not merely be retrieved. It should be explored.
Inspired by the ancient Greek maxim Γνῶθι σεαυτόν — Know thyself — SIBILA approaches artificial intelligence as a companion to human curiosity, not a substitute for human judgment.
Its guiding principle is simple:
SIBILA proposes. You decide.
What it does
SIBILA transforms a user-selected collection of NAS documents into an explainable, navigable atlas of ideas.
Its experience follows three natural movements:
- Discover: identify meaningful concepts within documents, respecting their individual contexts.
- Connect: explore possible relationships, supported by source evidence and clearly distinguished from interpretations.
- Explore: follow personally chosen directions, reconsider earlier connections, and build a traceable intellectual journey.
Imagine two readers exploring the same collection. One approaches Gothic culture through architecture, light, and symbolism. Another approaches it through literature, fear, and the unknown.
SIBILA is designed to support both journeys without imposing either.
It does not decide what deserves the reader's attention. It helps reveal what may be found along the path the reader chooses.
How we built it
SIBILA is designed as a native Uliya Skill for the UGREEN AI NAS environment.
Its Markdown-based SKILL.md orchestrates the platform's filesystem, document, and knowledge experts. The design requires no additional language model, external inference API, or independent database.
The architecture emphasizes explicit folder authorization, contextual concept identification, source-grounded relationships, and deliberate user choices.
We developed a deterministic testing laboratory using JavaScript, Node.js, Python, and synthetic PDF collections, with automated validation through GitHub Actions.
The current implementation passes 122 Node.js tests, alongside Python and OCR-related validation. These establish structural correctness and reproducibility, not yet native semantic performance on Uliya.
Challenges we faced
The greatest challenge was not discovering similarities. It was determining when a similarity deserves to become a connection.
A shared word does not necessarily imply a shared meaning. Two texts may explore similar themes without demonstrating historical influence.
SIBILA therefore distinguishes documented relationships, interpretative parallels, and insufficiently supported hypotheses.
Another challenge was preserving genuine intellectual freedom: allowing readers to deepen a subject, change perspective, reject suggestions, or return to earlier ideas without being pushed toward an algorithmically preferred destination.
Accomplishments and lessons learned
We have established an integrated Skill architecture, a source-evidence review contract, a conversational exploration model, and a reproducible validation process.
Our most important lesson is that an AI system becomes more trustworthy not when it always provides an answer, but when it knows what its evidence cannot justify.
SIBILA is currently statically validated and prepared for native-device testing; its real behavior on Uliya remains to be evaluated.
What's next
If selected as a finalist, we will deploy SIBILA on the UGREEN AI NAS, evaluate its interactions with real documents, test its contextual reasoning and source attribution, and refine the experience using observed results.
Our long-term vision is a personal knowledge atlas that can grow alongside its reader while keeping evidence transparent and intellectual direction firmly in human hands.
SIBILA proposes. You decide.
Built With
- github-actions
- javascript
- markdown
- node.js
- ocr
- pymupdf
- python
- tesseract
- ugos
- uliya