Nov 21, 2025
|
5
min read
Seed Week and Vibeathon at Bversity became much more than just another academic event, it turned into a real test of teamwork, clarity and problem solving. What started as a confusing challenge slowly transformed into a project that pushed our three learners to think, build and grow together. From collecting oncology datasets to shaping a knowledge graph and finally bringing an AI assistant to life, every step taught something new. This blog takes you through the OncoGraph journey, told in three voices across different phases. It is a story of how curiosity, patience and a few well timed insights can turn chaos into something meaningful.
“Challenges revealed the skills I never knew I had.”
Working on the OncoGraph: Biomarker Knowledge Navigator was one of the most insightful projects we have worked on, blending biomedical understanding with full-stack development. Since oncology information is extremely fragmented across multiple sources, manually gathering reliable data on genes, proteins, biomarkers, and cancer types would have been highly time-consuming. To save time while maintaining quality, our team used AI to generate a structured Excel file containing fields such as genes, gene type, protein, biomarker, drug, and cancer type and then manually verified every entry to ensure scientific accuracy and consistency. I converted this Excel file into CSV and then into a JSON structure, transforming each row into interconnected nodes and edges that formed the foundation of my knowledge graph. This JSON-powered graph enabled the frontend to clearly represent biological relationships like “BRCA1 → BRCA1 mutation → Olaparib → High Grade Serous Ovarian Cancer.”
“In a system where failure was allowed, success became inevitable.”
We then vibecoded the application in VS Code with GitHub Copilot using a three-stage approach - first, building a minimal web app that simply returned “Hello World”, then connecting it to OpenAI’s GPT-5 mini to generate actual responses and finally, integrating the knowledge graph and context file for our reference-focused query responses.
We did make mistakes. Our first attempt involved asking Copilot to generate the entire codebase from one giant prompt - a predictable failure. And while our final app was small- scaled and limited in handling varied query formats, we still presented it. My experience was defined by the supportive environment created by the Bversity team, Agilisium team, my batchmates and teammates - one where failing was allowed, and trying was encouraged. Within an intense 8-hour Vibeathon, that mindset made all the difference.
“Turning complexity into clarity, without losing what makes the vision powerful.”
At the Vibeathon, our team built Oncograph, a web application platform, born from a problem we couldn't ignore: why should researchers waste hours jumping between databases just to track down every piece of biological information they need? We wanted to change that. Our solution revolved around constructing a robust knowledge graph and enabling multi-hop querying, allowing users to fetch precise, context-rich answers directly through the application’s interface. Data reliability was a top priority, so we invested significant effort in carefully curating and validating datasets to ensure the platform remained grounded in accuracy. The experience pushed us technically and creatively, and the reward made it all worthwhile, reaffirming the strength of tight-knit teamwork and clear communication in high-pressure moments. Presenting our approach to the CEO and the Agilisium leadership taught me the importance of clear articulation, translating complex ideas for audiences with varied technical backgrounds while keeping our vision both accessible and deep!
