grounded-ai

A minimal graph engine for grounded AI — records, associates, and retrieves, but never invents. Written in Rust.

Agent Capability Standard is an open specification for composable AI agent capabilities. It defines 36 atomic capabilities across 9 cognitive layers, a type-safe workflow DSL, and grounded world modeling with trust-aware conflict resolution. Built on the Grounded Agency philosophy, it makes agent reliability structural—not optional.

DeepVault Nexus is a local-first RAG platform for administering governed knowledge sources and building source-grounded AI experiences.

Self-RAG chatbot with LangGraph that retrieves from internal PDFs, verifies grounding (IsSUP), checks usefulness (IsUSE), and iteratively rewrites queries to improve answer quality.

Corrective RAG with LangGraph: evaluates retrieval quality, routes to web search when needed, refines context, and generates grounded answers.

Agentic entity search engine that converts natural language queries into source-grounded entity tables using web search, scraping, LLM extraction, evidence verification, deduplication, ranking, and observability.

Agent-native digital ink SDK and protocol stack for structured, editable handwriting: stable strokes, deterministic analysis, bounded context retrieval, validated editing tools, UIM interoperability, and Python/C++ native hosts.

The Autonomous, Grounded Hardware Diagnostic Engine. A multi-modal infrastructure for physical-world engineering.

Instruction-following vision-language model (VLM): grounded text instructions executed via multi-modal reasoning

Grounded biomedical research agent. Six specialists investigate a target across any therapeutic modality, cite every claim to a real record (PMID, NCT, patent no.), verify it with a decorrelated model, and report the questions they could not answer. Abstains rather than guess. Runs locally on Ollama.

The Offline-Libby repository is a fully offline, air-gapped AI knowledge assistant designed for high-integrity information retrieval. It allows users to turn local document folders into a private, conversational knowledge base without any cloud dependency.

Local-first Python desktop application for coin and banknote collection management, analytics, OCR review and grounded AI.

Learning artifacts → learning gain, class gaps, and competency evidence (Bloom's).

QLoRA fine-tuning pipeline for multilingual grounded QA with Qwen3-4B

Evidence-grounded multi-agent vacation planner with parallel research, independent scoring, and verification to prevent source mixing.

Dense PDFs → coordinate-aware text, retrieval-ready chunks, and structured tasks.

A runnable reference core for the rule behind AI NCP: an AI system may retrieve and answer only from evidence the current person is entitled to access. Written for this repository, not lifted from the product. 9 tests, zero runtime dependencies.

Learner 'how do I learn?' survey → per-learner and per-class lesson-planning guidance.

Course JSON → a SCORM 1.2/2004 cartridge that drops into any LMS and reports scores.

Multi-agent AI business consultant that orchestrates specialised Finance, Marketing, HR, and Strategy agents to analyse business questions, validate recommendations, and deliver grounded strategic decision support through a React UI.