Developer creates open-source TRIZ-based AI agent for solving engineering problems

A software engineer has introduced the open-source ARIZ-agent project, combining classical inventive problem-solving methodology (TRIZ/ARIZ-85V), modern language models, and semantic search across patent databases.

Why automate inventive problem solving

The core idea of the project is to reduce the time needed to analyze technical problems and find out-of-the-box solutions. The user describes an engineering challenge, and the system breaks it down according to ARIZ algorithm rules, generates a spectrum of options with varying novelty levels, and immediately retrieves relevant patents from a vector store.

Architecture and tech stack

The solution is built on a combination of open-source and easy-to-deploy tools:

• n8n — handles orchestration and request processing pipeline logic;
• OpenWebUI — user chat interface triggering n8n webhooks via a custom function;
• Qdrant — vector database for storing and searching fragments of patent documents;
• GigaChat — language model serving as the agent's reasoning core;
• Patent Service — microservice for uploading and indexing data dumps from patent databases (such as Google Patents).

How to deploy the system

The project is designed for quick containerized deployment and is available in the GitHub repository.

1. Cloning and launch: clone the repository, prepare the .env environment configuration file, and start services via Docker Compose.
2. Configuring n8n: install the GigaChat integration node, add the API key to the Credentials section, and import the pre-built ARIZ workflow.
3. Connecting the interface: in OpenWebUI, disable automatic generation of system tags and configure the n8n webhook call.
4. Patent indexing: upload thematic CSV files via the patent service API into the Qdrant database.

Results and open-source access

After the problem description is submitted, the system produces a structured response ranging from conservative to radical concepts for developing the technical system, accompanied by patent references.

The project is distributed freely. The author notes that the modular architecture makes it possible to modify ARIZ workflows for custom tasks, as well as adapt integrations for other LLMs and knowledge bases.