Zendesk plugin and model tuner designed to handle customer support peaks with machine learning

Zendesk plugin designed to handle peak customer support volumes. Project in collaboration with bonsai.tech, a data science and ML lab, on a Zendesk plugin for customer support. Ticketflip uses machine learning and smart defaults to assign support tickets to the agent best equipped to handle each request. It was designed to seamlessly integrate into the Zendesk ecosystem and workflows

  • Client: bonsai.tech
  • Role: Product and design strategy, interaction design, UX design
  • Objective: Handle customer support peaks with ML-powered ticket assignment

Multi-modal context selector tool for AI agents and agentic IDEs. A scalable solution for persistent usability issue.

Premises: With the proliferation of AI agents and agentic IDEs, new usability challenges arise. Extensive prompting is not really a great experience, and describing the desired element can sometimes be almost impossible. Using auto-suggest and natural-language input in combination with a manual selector offers a flexible solution to this problem.

  • Objective: Extensive prompting is bad usability. HMW work around it?
  • Role: Research, concept development, prototyping, frontend development

Remix with AI - applied AI for music. An experiment with new musical tools and techniques.

Premises: Throughout the history of music, experimenting with new tools, techniques, and sounds was more about expression and finding the soul of music than about productivity. It's about using a contextual AI helper to bounce ideas and find new inspiration.

  • Objective: HMW make use AI as a creative partner instead of plagiarism machine
  • Role: Research, concept development, prototyping, UX design, UI design

Natural language to automation steps to CD/CI pipeline. Deterministic procedures as a guideline for probabilistic systems.

  • Case study: Coming soon...
  • Objective: HMW translate natural language into reliable automation
  • Role: Research, concept development, prototyping, frontend development

Pre-production screenshots: ML platform for developing investment strategies. Research, trading & risk management.

The platform provided fund-prediction signals and trading support to internal teams. Built as an analyst-first tool, it enabled analysts, fund and risk managers to monitor signals, run simulations and manage executions with clearly separated controls and cross-functional views. Features include a real-time signals overview, portfolio simulation, role-based dashboards, and a notification system. All built around four-level coexisting information architecture to surface context at glance, and enable cross-role coordination.

  • Note: Project from 2017, screenshots are pre-production version, UI was fixed in code
  • Client: Internal / proprietary (financial services)
  • Objective: Discover technical capabilities of early ML models, define roadmap and design scalable solution for predictive financial analysis. ML signals enable faster, data-driven decisions while maintaining rigorous risk controls and auditability
  • Role: Research, roadmapping, information architecture, UI design, UX design, frontend development