RYCO Job Opportunities Assistant
A chatbot prototype built to surface job listings across all six Western Balkan countries through a conversational interface instead of static listings pages, initially implemented on Microsoft Copilot and later rebuilt on the Grok LLM after a structured evaluation of response quality and flexibility.
Overview
A chatbot prototype for RYCO (Regional Youth Cooperation Office), built to surface job listings across all six Western Balkan member countries through a conversational interface instead of static listings pages.
Problem
Job listings and opportunities from RYCO's six Western Balkan member countries were scattered across static pages, making it slow for users to find relevant, current openings.
Context
Developed during my role as Technology Officer at RYCO, as part of broader efforts to modernize internal digital tools and improve how opportunities reach applicants across the region. This is a separate, earlier-stage prototype from the IT-support agent demo — the two address different problems and are not the same system.
My Role
I designed and prototyped the chatbot: an initial Microsoft Copilot-based implementation, then an evaluation of the Grok LLM as an alternative and a rebuild of the response layer on Grok after that evaluation.
Requirements
- Conversational interface for surfacing job listings across six countries
- Responses grounded in actual, current listing data
- Modern, approachable chat experience
Architecture
A frontend chat interface communicating with an LLM-driven response layer, initially implemented on Microsoft Copilot and rebuilt on the Grok LLM after evaluation.
Implementation
Built a chat-style interface for natural interaction with job-listing data spanning RYCO's six member countries. The first implementation used Microsoft Copilot; after evaluating response quality and flexibility, the response layer was rebuilt on the Grok LLM.
Technical Challenges
Evaluating Microsoft Copilot against the Grok LLM for response quality and flexibility, and migrating the response layer from one to the other, was the main technical challenge — alongside keeping listing data current across six distinct national sources.
Solutions
Ran a structured evaluation of Copilot versus Grok on response quality for this use case, then rebuilt the response layer on Grok once it proved the better fit for this data-grounding requirement.
Results
Evaluation is currently in progress. The Copilot-based prototype was replaced with a Grok LLM-based version after the response-quality evaluation; a public demo, deployment and adoption data have not yet been produced for this project.
Lessons Learned
Comparing two LLM providers on the same real use case made the practical trade-offs (response quality, cost, flexibility) concrete in a way that reading documentation alone doesn't.
Future Improvements
- Expand coverage to additional opportunity types beyond job listings
- Add multilingual support across the six countries
- Package a deployable demo once the Grok-based response layer is finalized
