3/9/2026
Scaling Pedagogical Pre-Training: From Optimal Mixing to 10B Tokens
Research on scaling pedagogical pre-training with optimal data mixing ratios, achieving improved perplexity on 10B token models with Sutra Framework.
3/9/2026
Show HN: Satellite imagery object detection using text prompts
Browser-based satellite imagery object detection using vision-language models with text prompts, outputs as GeoJSON.
3/9/2026
Show HN: I extended Shannon entropy with a time-dependent learning observer
Repository extending Shannon entropy with time-dependent learning observer for public research and audit, under restrictive source-available license.
3/9/2026
Show HN: Sentinel – Open-source MCP security scanner (config, probe, container)
Open-source CLI and GitHub Action for scanning MCP server security misconfigurations across configs, endpoints, and containers.
3/9/2026
Show HN: Run 500B+ Parameter LLMs Locally on a Mac Mini
OpenGraviton open-source inference engine running 500B+ parameter LLMs locally via ternary quantization and dynamic sparsity.
3/9/2026
LightReach: OpenAI gateway for Cursor(prompt compression+cost-aware routing)
OpenAI-compatible gateway with prompt compression, cost-aware routing, and budget enforcement for AI teams using Cursor.
3/9/2026
Show HN: Explorable World as Agent Skill
Bevy-based tool for AI agents to build 3D explorable worlds and save them as reusable skills in skill directories.
3/9/2026
Phi Guard – Open-Source HIPAA Phi Scanner for CI/CD Pipelines
Phi Guard is an open-source HIPAA PHI scanner for CI/CD pipelines that detects Protected Health Information in code repositories, complementing GitHub's Secret Scanning.
3/9/2026
ANIP: An agent-native protocol with REST, GraphQL, and MCP adapters
ANIP is a protocol standard enabling agents to reason before acting when calling APIs, with REST, GraphQL, and MCP adapters for agent-native interface design.
3/9/2026
Product Needs to Be a Runtime for Agents – Be a Container, Not a Wrapper
Hopsworks platform as runtime for coding agents like Claude Code, discusses agent infrastructure and data platform integration.
3/9/2026
Claude tested everything except the one thing that mattered
Case study of Claude Code building and testing a social application, demonstrating end-to-end test generation capabilities.
3/9/2026
Show HN: MCP Security Checklist – security controls for MCP server deployments
Community-maintained security baseline checklist for Model Context Protocol server deployments and AI agent infrastructure.
3/9/2026
Show HN: AI image models hallucinate history, we built a method to fix it it
Method injecting structured domain knowledge into image generation prompts to reduce historical hallucinations in AI image models.
3/9/2026
Show HN: Mcp2cli – One CLI for every API, 96-99% fewer tokens than native MCP
CLI tool converting MCP servers to command-line interfaces, reducing LLM context token usage by 96-99% through on-demand tool discovery.
3/9/2026
Show HN: Entropick – Plug quantum/hardware RNGs into LLM token sampling
entropick tool for replacing software PRNG with physical randomness in LLM token sampling, supports vLLM and Transformers.
3/9/2026
Show HN: Claude Code hook that nudges about accumulating WIP
AI agent hook for Claude Code providing visibility into accumulated work-in-progress like uncommitted changes and open PRs.
3/9/2026
The Audacious Roadmap for ADK-Rust
ADK-Rust announces roadmap to build Rust-native platform for building and deploying production AI agents across cloud, edge, and enterprise environments.
3/9/2026
Sandvault – Run AI agents isolated in a sandboxed macOS user account
SandVault is a CLI tool for macOS that sandboxes AI agents and shell commands in isolated user accounts, providing lightweight alternative to VMs.
3/9/2026
RoboLayout: Differentiable 3D Scene Generation for Embodied Agents
Differentiable 3D scene generation from language using vision language models for physically feasible embodied agent interactions.
3/9/2026
Real-Time AI Service Economy: A Framework for Agentic Computing Across the Continuum
Framework for deploying real-time AI services with autonomous agents across device-edge-cloud infrastructure with resource constraints.
3/9/2026
Reasoning Models Struggle to Control their Chains of Thought
Evaluation suite measuring whether reasoning models can control their chain-of-thought outputs, testing CoT controllability.
3/9/2026
Evolving Medical Imaging Agents via Experience-driven Self-skill Discovery
Medical imaging agents that self-discover skills and adapt tool invocation strategies through experience in clinical workflows.
3/9/2026
The World Won't Stay Still: Programmable Evolution for Agent Benchmarks
Framework for evolving agent benchmarks with dynamic environments and changing toolsets to test robustness of LLM-powered agents.
3/9/2026
DeepFact: Co-Evolving Benchmarks and Agents for Deep Research Factuality
Benchmark and co-evolving agents for verifying factuality in deep research reports generated by search-augmented LLM agents.
3/9/2026
An Interactive Multi-Agent System for Evaluation of New Product Concepts
Multi-agent system using LLMs to automate product concept evaluation by analyzing and scoring concepts across criteria.
3/9/2026
Agentic LLM Planning via Step-Wise PDDL Simulation: An Empirical Characterisation
Open-source PyPDDLEngine for LLM-based task planning using PDDL simulation, enabling agents to sequence actions via tool calls.
3/9/2026
Conversational Demand Response: Bidirectional Aggregator-Prosumer Coordination through Agentic AI
Conversational Demand Response mechanism enables bidirectional aggregator-prosumer coordination through agentic AI in energy systems.
3/9/2026
The EpisTwin: A Knowledge Graph-Grounded Neuro-Symbolic Architecture for Personal AI
EpisTwin neuro-symbolic architecture grounds personal AI reasoning in knowledge graphs and user-centric data for holistic sensemaking.
3/9/2026
SAHOO: Safeguarded Alignment for High-Order Optimization Objectives in Recursive Self-Improvement
SAHOO framework monitors alignment drift in recursive self-improving systems via Goal Drift Index and constraint preservation safeguards.
3/9/2026
Talk Freely, Execute Strictly: Schema-Gated Agentic AI for Flexible and Reproducible Scientific Workflows
Schema-gated agentic AI framework enables conversational flexibility while enforcing deterministic, auditable scientific workflow execution.
3/9/2026
Boosting deep Reinforcement Learning using pretraining with Logical Options
Hybrid two-stage framework combines logical options pretraining with deep reinforcement learning for aligned agent training.
3/9/2026
Can LLM Aid in Solving Constraints with Inductive Definitions?
LLMs with structured prompts generate auxiliary lemmas for constraint solving with inductive definitions, outperforming SMT/CHC solvers.
3/9/2026
Exploring Human-in-the-Loop Themes in AI Application Development: An Empirical Thematic Analysis
Empirical analysis of human-in-the-loop principles in AI application development, documenting organizational challenges and operational guidance.
3/9/2026
An Embodied Companion for Visual Storytelling
Companion artistic apparatus integrates drawing robot with LLMs for human-machine collaborative co-creation in visual storytelling.
3/9/2026
From Toil to Thought: Designing for Strategic Exploration and Responsible AI in Systematic Literature Reviews
Design study of systematic literature review tools identifies friction points in iterative scholarly work with AI-assisted exploration.
3/9/2026
Traversal-as-Policy: Log-Distilled Gated Behavior Trees as Externalized, Verifiable Policies for Safe, Robust, and Efficient Agents
Traversal-as-Policy distills LLM agent execution logs into Gated Behavior Trees for safe, verifiable, and robust agentic control.
3/9/2026
Molecular Representations for AI in Chemistry and Materials Science: An NLP Perspective
Survey of molecular representations for AI in chemistry from NLP perspective, comparing machine-readable and scientist-understandable formats.
3/9/2026
Omni-C: Compressing Heterogeneous Modalities into a Single Dense Encoder
Omni-C single dense Transformer encoder compresses heterogeneous multimodal inputs without separate expert encoders or routing overhead.
3/9/2026
Towards Neural Graph Data Management
NGDBench unified benchmark for neural graph database capabilities across finance, medicine, and AI agent tooling domains.
3/9/2026
VDCook:DIY video data cook your MLLMs
VDCook self-evolving video data platform generates multimodal datasets for MLLMs via natural language queries with real retrieval and synthesis modules.
3/9/2026
Human-Data Interaction, Exploration, and Visualization in the AI Era: Challenges and Opportunities
Survey of human-data interaction and visual analytics challenges with foundation models, LLMs, and VLMs for large-scale heterogeneous data analysis.
3/9/2026
EigenData: A Self-Evolving Multi-Agent Platform for Function-Calling Data Synthesis, Auditing, and Repair
EigenData platform automates function-calling agent training data lifecycle through multi-agent orchestration for tool-use LLM synthesis and auditing.
3/9/2026
CBR-to-SQL: Rethinking Retrieval-based Text-to-SQL using Case-based Reasoning in the Healthcare Domain
Case-based reasoning approach for LLM-to-SQL translation in healthcare domain, improving RAG for EHR database queries.
3/9/2026
PRISM: Personalized Refinement of Imitation Skills for Manipulation via Human Instructions
Instruction-conditioned method combining imitation learning and reinforcement learning for robotic manipulation refinement.
3/9/2026
Tool-Genesis: A Task-Driven Tool Creation Benchmark for Self-Evolving Language Agent
Benchmark for evaluating self-evolving language agents' capability to autonomously create and adapt tools from task requirements.
3/9/2026
DreamCAD: Scaling Multi-modal CAD Generation using Differentiable Parametric Surfaces
Multi-modal generative framework for CAD design using differentiable parametric surfaces and unannotated 3D mesh data.
3/9/2026
RACAS: Controlling Diverse Robots With a Single Agentic System
Multi-agent system enabling high-level autonomous control across diverse robot platforms via unified agentic interface.
3/9/2026
Relational Semantic Reasoning on 3D Scene Graphs for Open World Interactive Object Search
SCOUT system combines 3D scene graphs with semantic reasoning for efficient interactive object search without slow LLM inference.
3/9/2026
The Fragility Of Moral Judgment In Large Language Models
Framework testing stability and manipulability of LLM moral judgments using perturbation analysis on ethical dilemmas.
3/9/2026
SecureRAG-RTL: A Retrieval-Augmented, Multi-Agent, Zero-Shot LLM-Driven Framework for Hardware Vulnerability Detection
Multi-agent LLM framework with RAG for detecting hardware vulnerabilities in HDL designs, addressing knowledge gaps in security verification.