A unified encoder for safety moderation and PII detection in a single forward pass, built on top of GLiNER2.📄 Paper · 🤗 ModelsGuardrails in an agentic LLM application5 classifiers × 4 hops = 20 forward passes per requestSame application after GLiNER Gu…
System Architecture & Data PipelineWhen a customer types “something professional I can wear to a client meeting in summer” into a search bar, most e-commerce engines return zero results. Not because the products don’t exist — they do. It’s because …
SBERT, also known as Sentence-Bert, is a widely used approach for obtaining sentence embeddings that aim to retain the contextual information within the sentences. However, generating these embeddings can be slow when dealing with large amounts of data…
IntroductionContinue reading on Medium »
How to choose the right second-stage ranking layer for RAG when retrieval is good enough to find the answer but not good enough to…Continue reading on Towards AI »
A technical walkthrough of an AI-powered symptom analyzer with 7-language support, voice I/O, a dual-engine fallback architecture and a feature that guides users to nearby care based on where they actually are.Disclaimer: This project is for educationa…
Artificial Intelligence is rapidly transforming the way people interact with information. From virtual assistants and recommendation…Continue reading on Medium »
這篇是我在學習Nvidia: Building Transformer-Based Natural Language Processing Applications的筆記之一，如有錯誤請多多見諒&…
source: OpenAI GPT Image 2 modelYou have spent three weeks tuning hyperparameters, experimenting with the latest transformer architecture, and building a slick training pipeline. Your model is finally converging. But when you run inference on real data…
A group of Indigenous scholars is reimagining the foundation of AI principles.Continue reading on Medium »