Hacker News new | ask | show | jobs
Show HN: OntoCast – ontology-assisted KG generation (github.com)
9 points by acrostoic 360 days ago
Hey HN,

I'm excited to announce a new release of OntoCast — an open-source framework for extracting semantic triples and building knowledge graphs (KG) from unstructured documents (PDF, JSON, Markdown, and more).

Before extracting facts, OntoCast automatically selects or creates a relevant ontology and iteratively refines it, leading to much more accurate and context-aware fact extraction. This is especially valuable for cross-domain or complex documents where a static ontology falls short.

- Agentic workflow: Uses LLMs (OpenAI/Ollama) to drive the extraction and ontology refinement process.

- MCP-compatible API server: Easy to integrate into your stack.

- Flexible storage: Works with Jena Fuseki and Neo4j for knowledge graph storage.

- Open source: Apache licensed.

Uses cases include extracting structured knowledge from scientific papers, financial reports, or clinical trial documents — even when they span multiple domains.

Repo: https://github.com/growgraph/ontocast Docs: https://growgraph.github.io/ontocast

Would love feedback, questions, or suggestions!

2 comments

Been building several of these functionalities myself for a while... Happy to know someone more skilled did it also and released it publicly.
thank you! hopefully you will find it useful
Does this require to have any subscriptions to any LLM APIs? Thanks
thanks! Ontocast currently supports openai (subscription) and ollama (self-hosted) APIs.

At current openai pricing for GPT-4.1 mini $0.4/1M tokens we expect the cost of processing of 100 pages of text to be in the range of $0.02-0.08

NB: small models (< 14b) available on ollama struggle with structured output in our experience.

Thanks. Good luck!