CST-499 Week 4

Onco-Logic Week 4: Integration

This week marked a major turning point in our capstone project. We wrapped core development on the NLP pipeline and moved into the final report and presentation phase. Our team has been working in parallel across three machine learning tracks, and we’re now integrating the best components into a unified system.

What I accomplished this week:

• Built the demo app for the NLP project using Streamlit. It showcases our ClinicalBERT-based pipeline for extracting structured data from pathology reports.

• Integrated model outputs from earlier work into the main Onco-Logic system. This included updating preprocessing steps and formatting for consistent results.

• Solved compute access issues by setting up SSH connections to Google Colab. A teammate also found that Colab now offers free access to Pro-level GPUs, which will help us run large models in PyTorch and TensorFlow.

• Started drafting the final report and presentation, contributing summaries for the NLP and classification projects.

Plan for next week:

We will finish writing and formatting the technical report, finalize visuals for the presentation, and build out the demo apps for the survival prediction models. We'll also conduct internal reviews to ensure that all components work smoothly and integrate well.

Challenges:

Compute access was the biggest roadblock, especially for resource-heavy models. We’ve mostly addressed this through Colab. The other challenge is aligning outputs from different modules, since each team member worked independently. At this stage, we do not need instructor assistance.

Looking ahead:

We’re close to completing a full system that covers structured clinical data, genomic inputs, and free-text analysis. Working independently has paid off. Now, our focus is on refining and presenting a complete, functional AI pipeline for oncology.