GQA

PerceptionBench is a benchmark for evaluating atomic visual perception in multimodal large language models (MLLMs). It isolates perception from reasoning and domain knowledge by defining ten atomic perceptual capabilities, including counting, depth, localization, OCR, and perception-related hallucination, based on an error taxonomy derived from failures across existing benchmarks. The released dataset contains 3,000 verified open-ended questions with short, uniquely determined answers; 1,800 are decomposed from failures in source benchmarks and 1,200 are newly authored using supplementary images. The repository provides the benchmark's code and links to its paper and dataset.

QuerySplat is an open-source implementation for predicting and rendering 3D Gaussian splats from multiple images of a scene while decoupling geometry and appearance representations. Its inference pipeline preprocesses custom images, predicts 3D Gaussians, and uses VGGT-Omega components for camera and depth prediction; optional test-time optimization can refine the result. It can export Gaussian PLY files, predicted input cameras, per-view depth and confidence data, and depth-derived point clouds. The repository requires Linux, an NVIDIA CUDA-capable GPU, and CUDA-enabled PyTorch.

QMD is an on-device command-line search engine for Markdown notes, meeting transcripts, documentation, knowledge bases, and other organized information. It combines BM25 full-text search, vector semantic search, and LLM reranking locally through node-llama-cpp and GGUF models. Its hybrid query pipeline expands the original query into lexical, vector, and HyDE variants, routes them to keyword or vector search, combines the results with Reciprocal Rank Fusion, and applies an LLM reranker to produce final ranked results. Collections can be indexed and enriched with hierarchical context; documents can be retrieved individually or in batches, and JSON and file-oriented output formats support agent workflows. QMD also exposes query, retrieval, batch retrieval, and index-status operations through an MCP server.

GeoLook is an open-source, self-hosted platform for end-to-end Generative Engine Optimization (GEO), focused on whether AI engines mention and cite a brand. Its workflow runs from status analysis through diagnosis, strategy, implementation tickets, execution, and verification. The platform samples answers from multiple AI engines and measures mention rate, rank, citation share, competitor visibility, and citation sources. It classifies question-level problems such as absence, low ranking, competitor dominance, and suspected negative mentions; stores browsable sample metadata; and allows human corrections to parser results that recompute metrics and persist across resampling. It can expand question banks using Baidu suggestions and Google autocomplete, with candidates requiring manual review. GeoLook audits sites through an Access → Orientation → Understanding → Quotability dependency chain, covering crawler access, extraction blocks, citation-channel gaps, and brand-fact or messaging inconsistencies. It also performs checks such as robots and WAF/CDN behavior, noindex headers, llms.txt links, hreflang, sitemaps, duplicate content, and passage-level quotability. The action stage produces structured implementation tickets with rationale, ownership, effort, timing, acceptance criteria, risk grades, progress tracking, and regression-based reopening; it also provides content planning, deployable llms.txt and JSON-LD assets, and client-facing reports and execution plans. Subsequent sampling rounds compare results before and after implementation.

Google Cloud Agent Observability provides visibility into AI agents’ execution paths, actions, and reasoning, with evaluation capabilities for measures such as accuracy and hallucination detection. It is documented as part of Google Cloud Observability.