text-to-cypher

Easily get started with Spring-AI to develop various AI applications, including TextToSQL and private data AI application development. In addition to these capabilities, Spring-AI also supports integration with several other advanced AI technologies and platforms such as DeepSeek, Azure, Ollama, Vector Databases, Function Calling, MCP and RAG.

Open Source Instruction Tuning LLM Training Library 2026

An LLM training library for instruction-tuning.

Anime & manga recommendation agent over a Neo4j knowledge graph built from MyAnimeList. Local LLMs (gpt-oss-20b via MLX) for text-to-Cypher with schema and value checks, vector search over synopses, a tool-using agent, a web UI and automatic evals. Code and prompts in Portuguese (pt-BR).

A multi-agent natural-language query system for SQL, Cypher, and cross-source reasoning.

Implementation, experiments, and evaluation for translating Natural Language questions into Neo4j Cypher queries.

An advanced multi-agent research assistant using LangGraph, a Knowledge Graph (Neo4j), and a re-ranking RAG pipeline to reason over scientific papers.

Generate validated Cypher queries from natural language text with LLMs

Final project for Graph Database course — medical knowledge graph using Neo4j with LLM Text-to-Cypher, Graph Analytics (PageRank, Louvain), FastRP embeddings, and Graph-Augmented RAG over 100 diseases and 1,612 nodes.

Early skeleton: provenance-tracked document knowledge graph schema, entity resolution and a read-only Cypher guard, designed for graph RAG vs vector RAG

Text-to-Cypher RAG over a Neo4j movie knowledge graph. Ask in plain English, Gemini writes the Cypher, LangGraph repairs rejected queries, and the app streams the query, rows and answer. Built with Next.js, LangGraph JS and Ant Design.

A retrieval-augmented Q&A agent: natural-language-to-Cypher over a Neo4j knowledge graph with query validators, hybrid dense/sparse/parent-child retrieval, Redis caching, and eight advanced RAG patterns implemented end to end.

Ask about Formula 1 history in plain English: LangGraph text-to-Cypher agent over a Neo4j graph of every Grand Prix since 1950, with self-correction, read-only guards and an execution-accuracy eval.

Transparent AI knowledge graphs from regulatory documents. Ask questions in plain English - see exactly which strategy answered, the Cypher query generated, confidence score, and source documents.

Fine-tuned SmolLM2-135M-Instruct with LoRA to generate Cypher queries from natural-language questions and graph schemas. Built the complete training and evaluation pipeline, compared full-prompt and completion-only objectives, and improved Token F1 from 27.4% to 33.4% with completion-only training.

Media-monitoring workbench: congressional trades, insider filings and options flow in a queryable knowledge graph.

A shared operational memory over a company's work data that AI agents can query and cite. Loads real Jira data into Postgres and Neo4j, then compares graph and SQL answers on multi hop questions. Work in progress.

Proof-of-concept lab measuring security weaknesses in self-hosted text-to-Cypher GraphRAG pipelines: prompt injection, secret leaks, validator bypass and database roles (RBAC). Neo4j + LangChain + Ollama, four local models, 18 logged experiments.