Last week, the interop.io office was full of vibe as we had our second Hackathon this year.
In three days, the teams were able to submit six working projects that can bring value to our users and colleagues. The projects were:
io.Configgity
Setting up platform environments in io.Connect Desktop is powerful, but requires editing JSON files, managing multiple environment configurations, and repeating manual steps across teams.
io.Configgity replaces manual JSON editing with an interactive configuration wizard. Users can visually define applications, customize themes with drag-and-drop controls, and deploy new environments in minutes instead of hours. Power users still have direct access to configuration files when needed, but both technical and non-technical users now have the same power to create and customize environments, which dramatically shortens time to value for clients.
ai.Insights
Contact center agents, traders and portfolio managers juggle multiple applications throughout their day, but nobody really knows which apps they use, in what order, or where time gets wasted. Without visibility into actual workflows, optimization is guesswork.
ai.Insights transforms raw OpenTelemetry data from io.Connect into visual workflow maps. It reconstructs end-to-end user journeys by clustering telemetry data and using AI to identify patterns. The result is a dashboard that shows exactly which applications agents use, how much time each step takes, and where bottlenecks exist.
The real value is moving from intuition to evidence. Instead of assuming where inefficiencies are, teams can see actual user behavior and get AI-driven recommendations for improvement.
Chatbot Documentation Assistant
Documentation exists, but finding the right answer at the right time is still friction. Users search through pages, ask in Slack, or open support tickets for questions that documentation already has answers to. Meanwhile, generic AI chatbots like ChatGPT or Claude hallucinate answers or point users to outdated content.
This project builds an AI-powered documentation assistant that’s grounded in interop.io’s actual documentation and resources. Using RAG (Retrieval Augmented Generation) and fine-tuning techniques, the bot only answers when it knows what to say. When it doesn’t know, it directs users to the right support channels instead of making things up.
The team experimented with multiple approaches: RAG, fine-tuning, a combination of the two with various models, temperature settings, embeddings and data manipulation. An automated evaluation script ranks answers across configurations to find what works best. The goal is to deploy this on community.interop.io and the documentation site where developers can actually use it.
io.Codebox
Reading, without practice, doesn’t teach you how APIs really work. You need to experiment, break things, and see results. But setting up a test environment just to try a few API calls creates friction that stops exploration before it starts.
Io.Codebox gives developers a “tinkerable” environment where they can test io.Connect Desktop APIs instantly. Users select capabilities from a prompt library, choose examples, and generate working code in a sandbox. Single-mode generates standalone examples. Multi-app mode creates multiple windows that execute IO APIs and communicate with each other, all from natural language prompts.
Developers can verify their understanding of APIs in real-time, modify generated code, and see results immediately. No environment setup required, just instant experimentation that accelerates learning and validates concepts before committing to implementation.
io.AutoPilot
Repetitive tasks eat time. Every week, the same sequence: navigate to Jira, export data, format it in Excel, prep an email in Outlook, write a description, attach the table. Multiply this across teams and the wasted hours add up fast.
Io.AutoPilot automates pre-configured workflows in web applications. The proof of concept tackles weekly report generation: it can control a browser, navigates through saved reports, exports and formats data, and generates email content with the logged time table attached.
The system combines a local LLM with browser automation and a web UI for building workflows. It’s designed for hands-off execution of repetitive tasks so people can focus on creative work instead of mechanical processes.
Workspace Copilot
End-users switch between applications constantly, manually copying data, opening the same sets of apps for similar tasks. These “swivel chair tasks” waste time, reduce efficiency, and introduce errors. Current automation can’t adapt to changing workflows or learn from user behavior in real-time.
Workspace Copilot monitors user behavior in the Island enterprise browser (navigation patterns, copy-paste events, tab switching) and uses this data to recommend optimized workspaces. A browser extension captures events, sends them to the backend, where AI analyzes the data to identify frequently used app combinations.
When patterns emerge, users get recommendations for pre-configured workspaces that match their workflow. They can select the apps and launch a workspace with restored data context. The innovation is making workspace creation intelligent and adaptive based on user behavior. Instead of manually building layouts, the system learns from actual usage and suggests configurations that match how people really work.
Hackathon winners will be at the Client Forum on Thursday (Oct 23) and present their projects in the afternoon. If you want to learn more about any project, meet the team, or find it practical and want to explore it in detail, just let us know, and we’ll be happy to organize a meeting.