Software Engineering Hacker News

The Wisdom of Quinn the Eskimo (Apple Developer Technical Support Engineer)

A thread discussing the technical insights and support provided by Apple Developer Technical Support Engineer Quinn the Eskimo.

AI/ML arXiv cs.AI

GRACE-RAG: Governed Retrieval Architecture for Canonical Evidence Synthesis, Enabling Lightweight Deployment in Closed-Domain Institutional Settings

Introduces GRACE-RAG, a graph-augmented RAG architecture designed for lightweight, self-hosted deployment in closed-domain institutional settings.

AI/ML arXiv cs.AI

Towards an automated AI-based framework for floor plan compliance checks for residential buildings

Proposes an AI-based framework for automating residential building floor plan compliance checks using LLMs and structured building graphs.

AI/ML arXiv cs.AI

Libra: Training the Environment for Agentic Information Retrieval

Presents Libra, a self-evolving framework that optimizes agentic information retrieval by creating mutable catalogs within repositories.

AI/ML arXiv cs.AI

Learning User-Aware Recall: Personalized Retrieval in Long-Term Conversational Memory

Introduces PPRO, a retrieval-centric framework for personalized long-term conversational memory in LLM agents.

AI/ML arXiv cs.AI

LLMs in the Real World: Evaluating "AI" in Emergency Contexts

A call to action urging researchers to be more transparent about LLM limitations, illustrated by a case study on emergency 911 translation systems.

AI/ML arXiv cs.AI

Aligning Sentence Embeddings to Human Concepts via Sparse Autoencoders

Proposes using Sparse Autoencoders to disentangle sentence embeddings for more interpretable and steerable RAG retrieval processes.

AI/ML arXiv cs.AI

FLYNN: Robust Neural Network for Robot Navigation using Fly Brain Topology

Introduces FLYNN, a neural network for robot navigation based on the biological brain topology of a fruit fly, demonstrating superior robustness to sensory loss.

AI/ML arXiv cs.AI

Memory-Native Non-Terrestrial Networks for Embodied Intelligence

Proposes MemNTN, a memory-native non-terrestrial network paradigm to improve connectivity for embodied intelligence in wilderness environments.

AI/ML Hacker News

CursorBench 3.1

A discussion or update regarding CursorBench 3.1, a benchmark for evaluating AI-powered code editors.

AI/ML arXiv cs.AI

Why Advanced Encoders Lag on Sparse Retrieval? The Answer and an Approach to Bridging Vocabulary Gaps

Introduces Vocabulary Transfer (VT), a framework to bridge vocabulary gaps in advanced encoders like ModernBERT to improve learned sparse retrieval performance.

AI/ML arXiv cs.AI

Topological Void Analysis A Mathematical Framework for Systematic Technical Innovation Discovery in Knowledge Spaces

Presents Topological Void Analysis (TVA), a mathematical framework for discovering unexplored innovation opportunities in dense technical knowledge spaces.

AI/ML arXiv cs.AI

Persona Without Substrate: Regime-Dependence and the LLM Individuation Problem

Explores the LLM individuation problem and proposes regime-indexed individuation to explain how persona-vectors vary across different operational regimes.

AI/ML arXiv cs.AI

BaRA: BFS-and-Reflection Web Data Collection Agent

Introduces BaRA, a web data collection agent combining BFS traversal and self-reflection to improve multimodal extraction and link discovery.

AI/ML arXiv cs.AI

SchemaRAG: Dynamic Large Schema Reduction for LLM-driven Structured Information Extraction

Proposes SchemaRAG, a RAG framework that dynamically prunes large output schemas to reduce latency and token costs in structured information extraction.

AI/ML arXiv cs.AI

Controllable Narrative Rendering for Enhanced Assisted Writing

Presents Loom, an assisted writing framework that uses a three-layer pipeline to provide precise control over narrative intent and rendering density.

AI/ML arXiv cs.AI

Prompt Optimization for User Simulation in Conversational Recommender Systems: A Multi-Objective Framework

Proposes a multi-objective framework for automatically optimizing prompts for LLM-based user simulators in conversational recommender systems.

AI/ML arXiv cs.AI

SkillSelect-Serve: Budget-Controllable and QoS-Aware Skill Service Recommendation and Composition for Small LLM Agents

Introduces SkillSelect-Serve, a budget-controllable framework for recommending and composing LLM agent skills based on QoS-related attributes.

AI/ML arXiv cs.AI

PRA-RAG: Provably Robust Aggregation in Retrieval-Augmented Generation against Retrieval Corruption

Introduces PRA-RAG, a robust retrieval aggregation algorithm designed to defend RAG systems against poisoning attacks using geometric embedding structures.

AI/ML Hacker News

Kimi K2.7 Code is generally available in GitHub Copilot

Kimi K2.7 Code, a coding-specific model, is now available for use within GitHub Copilot.