Bioscribe Lab Companion

From messy notes to verified science in seconds.

Comment

Inspiration

My inspiration behind Bioscribe, as a Biology tutor and computer engineering student, was that I noticed a critical feedback gap in scientific education. Students often wait days for feedback on their handwritten lab notes, frequently internalising major scientific misconceptions - like the common error of placing carbocation intermediates in concerted E2 reactions. I built Bioscribe to provide a real-time, expert-level tutor that lives right on the digital lab bench. Notes can go from paper to digital and can be saved and exported, AI driven insights can be generated from questions asked about the lab notes content and these insights can be challenged and verified.

What it does

Bioscribe is an intelligent laboratory companion which digitises, summarises and challenges handwritten notes and specimens. Using a dual-mode interface, it serves both primary school and university researchers. It doesn't just transcribe; it uses a reasoning critique loop to identify mechanistic errors, verify chemical accuracy and log findings into a secure lab ledger for future use.

How we built it

BioScribe is built entirely within the Snowflake Data Cloud:

  • Utilised Snowflake Cortex AI (Claude 3.7 Sonnet) for multimodal vision and scientific reasoning.

  • Backend: Used Snowpark Python to handle file streams and SQL execution natively in the warehouse.

  • Data lifecycle: Raw images are stored in Snowflake stages, while processed insights are persisted in a structured Snowflake table.

  • Frontend: A custom-themed Streamlit dashboard designed for accessibility and a professional lab aesthetic.

Challenges we ran into

The biggest challenge was handling multimodal hallucination. To ensure educational integrity, I implemented a second pass critique agent which cross-references the initial transcription against established biochemical laws. This ensures the app doesn't just repeat what the student wrote, but actively corrects them.

Accomplishments that we're proud of

I am proud of producing a full stack app on Snowflake using its capabilities, and that the app assists students from primary school to university with AI driven insights whilst challenging any inaccurate information and tailoring scientific communication to each audience to make science an exciting subject, which is textbook heavy and aims for breadth rather than depth. I am very happy to bridge my expertise in Bioinformatics and Computer Engineering to solve a real pedagogical problem.

What I learned

  • I learned to use a dual-pass AI system to catch complex scientific errors (like E2 carbocation myths) that a single pass would miss.
  • I learnt how to build zero-egress apps using Snowpark and Cortex, keeping sensitive student data secure within the warehouse.
  • I learned to manage the journey from raw pixels in Snowflake stages to structured insights in relational tables.
  • I learned to design a single UI that scales in complexity from primary school to university researchers using custom CSS and logic.

What's next for Bioscribe Lab Companion

  • Integrating Augmented Reality - Moving from 2D photos to 3D specimen analysis using mobile XR.

  • Integrating with professional laboratory information management systems for industry-grade data logging.

  • Offline support: Exploring edge-computing models for fieldwork in remote ecology studies.

Built With

Share this project:

Updates