AI EngineeringCompleted
Solide Query Distributor
A Java desktop app that makes internal information accessible through a local Ollama model, query logic and conversation management.
Solo developer & designer · 2024
- Java
- CSS
- Ollama
- Confluence
- Elasticsearch
Built: local Ollama model, query strategy and session management
View source code on GitHub ↗
Problem
Employees have plenty of internal information, but do not always have a fast, privacy-conscious way to ask specific questions about it.
Approach
A local Java application combining query logic, internal sources and Ollama, displaying answers within each conversation session.
Engineering decisions
- Using a local model to keep data within organizational boundaries and support offline use
- Embedding AI functionality in a desktop application with authentication and multi-tab session management
- Managing conversation state explicitly so multiple topics remain usable alongside one another
Read the technical details
SolideQueryDistributor explores how employees can find internal information faster without sending organizational data to an external model provider. The Java application connects application and query logic to internal sources such as Confluence and Elasticsearch, and uses Ollama to generate answers locally. User authentication, multi-tab sessions and an offline mode make the AI prototype part of a usable application rather than a standalone model demo.
AI component
The AI layer consists of a local Ollama model called from the application. The application uses its own QueryResolutionStrategy and connects internal sources such as Confluence and Elasticsearch to the query flow. This is an applied local LLM prototype, not a claim to a fully production-grade RAG platform.
Features
- AI query processing using a custom QueryResolutionStrategy and a local Ollama model
- Integration with internal sources such as Confluence and Elasticsearch
- Multi-tab session management with conversation history per topic
- Offline mode — fully functional without an internet connection
- Multilingual support and user authentication with profile management
Challenges
- Integrating a local LLM (Ollama) into a Java desktop application with a robust query strategy
- Providing offline functionality without sacrificing response quality
- Building an intuitive, multilingual GUI with session management for simultaneous conversations
Lessons learned
- How local AI models work and their trade-offs compared with cloud-based APIs
- Integrating internal sources (Confluence, Elasticsearch) for contextual answers
- The project won the 2024 Audience Award at The Hague University of Applied Sciences
The Hague University of Applied Sciences
Audience Award
Winner 2024SolideQueryDistributor won the 2024 Audience Award during the project presentations at The Hague University of Applied Sciences.