Team OverSlept

Team Members

Marsel Senka , George Souladakis

Track

Track 01 - Traditional Knowledge Mining

Inspiration

Historical medicinal texts contain valuable knowledge, but their plant names and medical terms are difficult to connect reliably to modern science. We wanted to make those connections easier to explore without hiding uncertainty.

What it does

EthnoTrace connects historical passages and medicinal claims to possible modern plant identities and selected research. Users can explore the evidence through an interactive knowledge graph and atlas.

How we built it

We used Python to turn historical texts into structured entries, preserving names, passages, descriptions, and medicinal claims.

Our historical sources include Culpeper’s The Complete Herbal, Dioscorides’ De materia medica, and a later reconstruction of the Shennong Bencao Jing. The editions, data sources, and licence information are documented in our GitHub repository.

To identify possible modern plants, we searched sources such as GBIF, Wikidata, and historical dictionaries. We then used World Flora Online and WCVP to check scientific names and taxonomic classifications, alongside geographical and descriptive evidence.

Google Gemini assisted with extracting claims, translating quotations, comparing botanical descriptions, and interpreting selected research. We checked supporting quotations against available source texts and used predefined rules to categorize plant candidates and assess evidence.

We also used OpenAI Codex and Anthropic Claude to assist with development, documentation, and reviews.

We connected the results in a Neo4j knowledge graph, then built a React and TypeScript interface with a Python API. The frontend is hosted on Vercel, the API on Render, and the graph on Neo4j Aura.

Challenges we ran into

Historical names can refer to multiple plants, scientific sources can disagree, and AI interpretations can contain errors. We worked to preserve quotations, competing candidates, and source provenance.

Accomplishments that we're proud of

We built a traceable dataset containing over 10,000 historical claims and an interface that connects them to proposed plant identities and research from various sources.

What we learned

We learned that structuring information is only the beginning. The difficult part is preserving what the information means, where it came from, and how uncertain each connection may be.

We also learned that AI can help process complex material, but it cannot replace domain expertise. A plausible interpretation or a high category in our pipeline is not the same as an independently verified conclusion.

Our team is not in a position to independently validate botanical, historical-medical, or clinical findings.

What's next for Team OverSlept

Our next priority is to work with botanists, historians of medicine, and other domain experts to review samples of the pipeline’s outputs and identify where the system needs improvement.

Explore the project

The prototype is intended for research exploration. Its outputs are not proof that a historical remedy works and should not be used as medical advice.