BrainBridge, a translator for scientific jargon
I worked on a tool that transcribes multidisciplinary discussions and explains scientific jargon in plain language.
Conceptual diagram; implementation details are simplified for clarity.
The problem
Multidisciplinary teams can lose the thread of a discussion when a technical term is unfamiliar. BrainBridge aims to make those conversations more accessible by transcribing discussion and translating scientific jargon into plain language.
What I owned
I helped design and build the end-to-end language pipeline, including the jargon-detection flow and retrieval-augmented generation architecture. I also generated and cleaned more than 5,000 text samples to support this work.
The technical approach
I designed the pipeline as two connected tasks: BERT identifies jargon in a transcript, then the RAG architecture retrieves relevant context to ground a plain-language explanation. I built the components that connect detection, retrieval, and generation.
A RAG-enhanced LLaMA approach was explored during the project; this description does not imply that a specific model was ultimately deployed.
- Transcribe a multidisciplinary discussion.
- Use BERT to identify unfamiliar scientific terminology.
- Explore retrieval-augmented context for plain-language explanations.
What changed
The project created a path from spoken technical discussion to contextual explanations, helping participants get the meaning of unfamiliar terms without derailing the conversation.