human-centered-ai

Agent Skills for working UX designers. Challenge ideas, find blind spots, use the real design system, preserve evidence and decisions, and carry UX intent into engineering.

What's In My Human Feedback? Explaining preferences in human feedback using interpretability + LLMs. https://arxiv.org/abs/2510.26202

Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States

🔥 [ICML 2026] Official implementation of "Are LRMs Interruptible?"

CognitiveLens is a Streamlit-powered analytics tool for exploring alignment between human and AI decisions. It visualizes fairness, calibration, and interpretability through metrics like Cohen’s κ, AUC, and Brier score. Designed for ethical AI, bias auditing, and decision transparency in machine learning systems.

An explanation-first HR analytics system that reconstructs why employee exit becomes rational. Instead of predicting attrition, it generates human-readable exit narratives by decomposing pressure and retention forces, adding peer context and counterfactual interventions to reveal how stability erodes over time.

An early-warning system that models disasters as instability transitions rather than isolated events. It combines force-based instability modeling with an interpretable ML escalation-risk layer to detect when hazards become disasters due to exposure growth, response delays, and buffer collapse.

A Unified Survey of Human–Object Relationship Understanding: From Gaze to Interaction

[AAAI 2026] Learning from Human Gaze: Human-like Robot Social Navigation in Dense Crowds

Repository of the paper "Toward a Responsible Fairness Analysis: From Binary to Multiclass and Multigroup Assessment in Graph Neural Network-Based User Modeling Tasks"

Emotion-aware AI companion designed for natural, human-like conversation. TalkSpace combines machine learning, LLMs, and real-time UI adaptation to create a calm, supportive space where users feel heard — not analysed 🌿

The Conscience Layer Prototype, created by Aleksandar Rodić in 2025, establishes a research foundation for ethical artificial intelligence. It brings moral awareness into computation through principles of truth, human autonomy, and societal responsibility, defining a transparent and accountable form of intelligence.

IEvoAgent: Evolving Conversational Agent based on User Implicit Feedback (ACL 2026 Oral). Leverages conditional feedback distribution matrix with offline KTO alignment and inference-time prompt evolution for multi-turn dialogue.

Human-centered multimodal sexism detection for EXIST 2026 (CLEF) Task 2: a five-stream FiLM-conditioned cross-attention model that fuses meme text, image, annotator demographics, and biometric signals. Team VANGUARD.

Local-first personal AI assistant with chat, memory, coaching, and multimodal explain workflows.

Experimental framework for evidence-grounded human capability modeling, governed evaluation, progression, and safe external observation.

Framework de Desarrollo para Modelo Etico Adaptativo

A safe and explainable multi agent AI assistant for elderly care, with routing, RAG, guardrails, and caregiver escalation.

Organizational Systems Intelligence (OSI) and Professional Identity Architecture (PIA): a human-centered, evidence-governed research and engineering project for capability, organizational systems, and trustworthy machine participation.

Agentic RAG system — routes natural language to SQL, industry documents, or both. LangGraph + Pinecone + GPT-4o-mini. 86.1% benchmark accuracy, 0% SQL hallucination.