Physics-constrained neural network for correcting indentation size effects in steel hardness measurements from shallow nanoindentation tests.
arXiv paper on model-based bootstrap for transition kernels in controlled Markov chains with applications to offline reinforcement learning.
arXiv research on semantic mutation testing metric for scientific computing programs addressing domain semantics.
arXiv research on distribution-free confidence set construction for changepoint localization after sequential detection.
arXiv paper proposing transparent synthetic data generation method with guarantees on variable relationships and privacy.
arXiv research on CARVE, recurrent model with improved state management for memory-aware chunk-parallel linear attention.
Study analyzing bird diversity across Sri Lanka using environmental data integration and spatial analysis.
arXiv paper on CONFLUX, latent diffusion model for 3D chest CT synthesis with reinforcement learning post-training.
arXiv paper on Wan-Streamer v0.2, audio-visual interaction model achieving higher resolution (640x368) while maintaining 200ms latency.
arXiv paper on TACTIC-KG using small LLM agent teams for constructing cybersecurity threat intelligence knowledge graphs from unstructured text.
arXiv research on learning physically-relevant information in high-dimensional dynamical systems analysis.
arXiv paper introducing Nemotron Audex-30B-A3B, unified audio-text LLM maintaining text performance while adding audio understanding.
arXiv research on topology-aware CNN approach for reducing false positives in intracranial aneurysm detection from medical imaging.
arXiv paper on multiplayer world models for dynamic environments with complex physics, conditioning on multiple agent action streams.
arXiv research introducing LLM-as-a-Verifier framework for verification as scaling axis, enabling fine-grained feedback for agentic tasks.
Traceburn profiler analyzes LLM agent token usage and identified 69% avoidable spending by recommending prompt caching optimization.
Go-to-market strategy for developer tool companies using engineering-driven systems instead of traditional marketing campaigns.
Technical analysis of Git commit signature malleability vulnerabilities and their real-world implications.
Smolbren converts markdown files into queryable knowledge graphs with full-text search, supporting AI agent queries via Cypher and BM25.
Desk-Pet: local-first desktop companion powered by MiniCPM5 LLM; fully offline with guided setup for macOS and Windows.
Activity-frames tool gives AI agents real-time visibility into user activity and calendar.
OCR Buddy: local browser OCR extension for code, formulas, tables with Markdown export for LLMs; fully offline, no hallucinations.
Analysis of Chinese AI models gaining market share in US as costs rise and performance improves.
Personal story about learning Lua and building a game, unrelated to AI/ML.
Backlog tool for managing tasks and context for AI coding agents.
Linux Foundation standardizing AI agent identification through Agent Name Service (ANS) and DNS-AID protocols.
Omnibaas: provider-agnostic Infrastructure-as-Code compiler for BaaS services reducing vendor lock-in across multiple cloud providers.
Hardware project: enthusiast built 8,192-core GPU from RISC-V microcontrollers as alternative to expensive commercial GPUs.
Research on GPU-accelerated SVD computation using mixed-precision iterative refinement for numerical stability.
IRC-style message bus infrastructure for AI agents using Model Context Protocol (MCP).
Comparison of Meta's Muse Image model against other image generation models.
CLI tool for curated AI news feed with source citations, alternative to LinkedIn AI coverage.
Analysis of code maintainability challenges from agentic AI coding tools violating DRY principles.
Windows desktop app scheduling PC shutdown with countdown timer and automatic trigger on download completion.
Workmux: workflow tool for managing git worktrees and tmux windows as isolated development environments, optimized for parallel AI agent execution.
Service offering code refactoring for AI-generated codebases to improve maintainability and reduce duplication.
Launch playbook framework for coding agents and agentic builders to ship products through community engagement without existing audience.
Analysis of sandbox escape vulnerability in Cursor IDE, highlights security needs for AI agents with system access.
OpenGenie: AI-powered game development agent converting plain language descriptions into Godot 4 projects with automated testing.
Free PDF document redaction tool claiming better reliability than GPT/Claude with SOC-2 certification.
GitHub account lockout complaint: user's repositories and profile disappeared without warning or human appeal process.
Microsoft replacing OpenAI and Anthropic models with proprietary AI in Excel and Outlook to reduce costs.
Claude's learning mode now available to all Code users, guiding users to solutions rather than direct answers.
Dejavu: PATH shim tool reducing redundant command output for coding agents like Claude Code and Cursor by returning compact deltas instead of repeated output.
MIRA: 5B parameter latent diffusion world model for Rocket League supporting real-time 2v2 multiplayer gameplay at 20 FPS on single GPU with released code.
CLRK: open-source agent runtime using gVisor for isolation and Kubernetes deployment. Provides fully intercepted I/O for tracing LLM and network calls with MitM guardrails.
Open source platform for AI interpretability research enabling analysis of neural network internal mechanisms.
Proxy tool that adds vision capabilities to text-only LLMs like GLM 5.2.
Q is a REPL interface for running LLMs directly within the terminal.
Tool to help AI agents on Cloudflare infrastructure.