Discusso was engineered to solve the "Noise Problem" in modern digital forums. Most platforms rank content based on simple engagement—upvotes and comments—which often rewards clickbait over substance. I built Discusso as a microservice-oriented platform that uses Machine Learning to evaluate the semantic quality of a post before it even reaches the feed. By calculating "Effort" and "Openness" scores using Transformer-based NLP models, the system ensures that high-quality, thoughtful contributions are prioritized, regardless of how many followers the author has.

The technical challenge involved orchestrating a seamless "Handshake" between a Next.js frontend and a FastAPI backend. To ensure production-grade reliability, I implemented a robust CI/CD pipeline using GitHub Actions and Docker. This pipeline enforces an 86% test coverage gate using Playwright for End-to-End verification. The system doesn't just check if the buttons work; it verifies that the ML model correctly categorizes and scores content in a containerized environment before any code is deployed to the live site.

On the backend, I prioritized architectural stability over "hype." While deep learning models are popular, I chose a hybrid approach using Sentence-Transformers with a Logistic Regression head. This decision was driven by the current dataset size, ensuring the model remains interpretable and avoids overfitting while maintaining an inference latency of under 150ms. The result is a platform that feels as fast as a traditional forum but possesses the hidden intelligence to foster meaningful, high-effort human dialogue.