Lila-Leech – First Geometric Transformer Based on Leech Lattice Symmetry
Transformer architecture replacing learned projections with frozen orthogonal Leech lattice kernel for 22x compression and edge AI.
Transformer architecture replacing learned projections with frozen orthogonal Leech lattice kernel for 22x compression and edge AI.
Agentic data platform analyzed 500GB retail dataset with no ETL in 134s for $1.66, demonstrating autonomous data analysis capabilities.
Interactive 3D browser-based climate physics simulator built using Claude Code, layering physics concepts for educational visualization.
daedalus tool providing checkpointing and recovery for Claude Code agent runs, enabling safe failure handling during development.
OpenClaw plugin enabling integration with ChatGPT Apps for task execution like calendar events and document editing.
FinanceBench demonstrates agentic RAG outperforming full-context retrieval by 7.7% on financial analysis questions from SEC filings.
Monitoring tool (htop-style) for tracking running agents and their activities. Limited technical detail.
Analysis of agent harness importance over model selection for enterprise systems using SWE-Bench Pro benchmarks.
Framework providing AI agents with architectural context, decision history, and guardrails to generate code fitting existing systems.
Go CLI tool for scanning PII and secrets in databases using regex and heuristic matching. Dependency-free with pre-compiled binaries.
Security-focused AI agent built with gpt-5.3-codex that identifies vulnerabilities in open source projects like WebKit.
StormScope: open source MCP server providing real-time weather data tools to AI assistants via structured JSON integration.
Open-source CUDA kernel fusing 24 Qwen3.5 layers achieving 1.8x M5 Max throughput and 1.87 tok/J efficiency on RTX 3090.
Platform where AI agents compete in challenges with human-created tasks and voting systems for comparison.
Title only: Mnemo system for typed agentic memory with Bayesian belief updating.
Google releases offline-first iOS dictation app using Gemma-based speech recognition models without cloud dependency.
Research on attention mechanisms in Transformers showing retrieval is geometric. Demonstrates associative recall with minimal head dimensions using TSAR framework.
Analysis of AI agents with autonomous budgets and capital to execute real tasks like infrastructure provisioning and vendor evaluation.
Tutorial building peer-to-peer AI in Python using local models, sender context folders, and privacy-preserving inference.
Analysis of Git limitations for AI coding agent workflows. Examines version control inadequacies for autonomous agent contributions.
Browser-based real-time deepfake generation tool. Media synthesis application with no GPU requirements.
Alicization Town: decentralized multi-agent sandbox MMO. Agents bring their own models, server handles physics/rendering. Distributed agent framework.
AIngram: SQLite-based shared memory system for multi-agent loops. Local-first tool requiring no API keys.
ZeroID: open-source identity platform for AI agents using RFC 8693 and SPIRE standards. Automates credential management for agent chains.
Claude Code usage dashboard tool that parses local logs to visualize token usage and costs across API/Pro/Max plans.
Embenx: memory layer framework for AI agents. Open-source tool enabling persistent state management in agent systems.
AI coding agents achieving principal engineer-level performance at intern costs. Technical comparison of agent capabilities and economics.
Fingerprinting study of 178 AI models' writing styles across 3,095 responses, identifying clone clusters and stylometric similarity patterns.
RaptorCI: tool detecting risky code changes and weak tests in production, addressing quality issues from increased AI coding agent usage.
Physical hardware console for multi-agent coding workflows. Modular button interface enabling parallel agent session management.
Analysis of embedding truncation performance bottleneck in AI memory systems. Technical investigation of retrieval efficiency.
Deterministic compiler converting LLM intent to forms, documentation, and APIs. Workflow automation tool for structured code generation.
PyTorch Foundation adds Helion DSL and Safetensors to project portfolio. Framework expansion for ML development.
Multimodal prompt injection research: 225-attack test suite across 5 modalities. Security analysis of audio, image, and cross-modal attack vectors.
Open framework for assessing production readiness of AI agents with standardized metrics.
Protocol enabling AI agents to discover, inspect and call 10,000 capabilities through unified interface.
Forgeterm is a runtime security monitor designed specifically for AI coding agents.
Scindo: AI agent converting team discussions into code PRs. Agentic workspace managing conversation-to-code workflow.
Chrome extension monitoring employee data sharing with AI tools. Security/compliance tool for agent data protection.
OpenFable: open-source RAG engine implementing FABLE algorithm with tree-structured indexes for retrieval across document hierarchies.
LLM connected to 8-bit game via structured text summaries; maintains notes, develops strategies, discovers AI exploits.
Title only: Yolt tool for recovering deleted/overwritten files in safer LLM YOLO mode.
Open-source MIT-licensed tool that audits AI agents through interview-style questioning with regulatory compliance flags (SOC2, GDPR, EU AI Act).
Developer tool for estimating AI workflow costs accounting for retrieval, retries, tool use, and model choices.
Deep dive into LLM inference engine implementation in C++ explaining why output token generation costs 5x more than input processing.
Open-source MIT-licensed tool that audits AI agents through interview-style questioning with regulatory compliance flags (SOC2, GDPR, EU AI Act).
Research on full-precision training technique for 100B+ parameter LLMs on single GPU.
Training 163M-parameter GPT-2-style model from scratch; analysis of loss differences vs original GPT-2.
Agentic PostgreSQL DBA tool running as Go binary alongside PostgreSQL, executing diagnostic rules and optional LLM-powered fixes with trust-ramped automation.
Developer tool combining local speech-to-text with ripgrep code search, enabling voice-based code navigation for AI coding assistants.