Ryan Neufeldt
Email me
← Back

Ryan Neufeldt

Conversation Designer · LLM Agents & Behaviour Rules · Edge Cases & Fallbacks · Evaluation & User Testing

Conversation designer for LLM agents: writes the prompts, tool rules and behaviour rules that decide what an assistant does next, and maps the edge cases and fallbacks behind them. Co-built Recharge's GenAI support agent with Claude Code, including its skills, tool calls, page-aware guidance and the evaluation loop that tested every change. Before that, 4+ years designing and running the Ada support bot, and 8 years in insurance, including Product Owner work. On my own time I build working apps, prototypes and presentations with Claude Code.

Recharge

Remote · Feb 2022 – Aug 2026

Support Bot Designer

Mar 2026 – Aug 2026

  • Co-built Remi, Recharge's generative AI customer support agent, with a small team using Claude Code; personally built its evaluation system (judges and golden sets) and the self-healing improvement loop, and improved its skills and tool and API calls
  • Designed a human-in-the-loop feedback system that lets the bot surface its own weak spots for review before issues reach merchants
  • Designed and implemented the page-level guidance framework, so the assistant knows what the merchant is looking at and can often answer a pasted screenshot with no typed words
  • Designed and ran the go/no-go benchmark for Remi: a head-to-head against Ada, the live third-party bot, on a set of real merchant questions, scored with a rubric for correctness and completeness, followed by a staged rollout to a test group of merchants
  • Engineered self-healing automation using gradual, iterative fixes informed by real data flows to raise accuracy without introducing new risk

Knowledge Specialist / Bot Manager

Feb 2022 – Mar 2026

  • Owned end-to-end strategy, onboarding, launch, and ongoing optimization of Recharge's Ada-powered support bot (4+ years)
  • Partnered with Product, Engineering, and CX to design and configure logic-based workflows, decision trees, and conditional automation
  • Redesigned the live-agent handoff from five steps to one confirmation question, with an auto-generated summary, store context and a priority rubric; iterated from real conversations until nearly everyone completed it without repeating themselves
  • Tracked emerging support trends and translated them into new playbooks and conditional flows, continuously raising intent coverage
  • Built carousel-style choices with images that the bot showed in place of a text reply when the question type called for it
  • Released flow changes to a subset of users first, then refined them before rolling out to larger groups
  • Maintained bot performance reporting and roadmap; communicated data findings to VP-level stakeholders
  • Developed AI-ready knowledge content strategies, restructuring source content for AI retrieval and accuracy

Manulife

Kitchener, ON · 2011 – 2019

Digital Claims Agile Product Owner

2016 – 2019

  • During the requirements phase of the call center's IVR build, gathered user journeys and mapped the caller's flow through the system
  • Led implementation and redesign of a legacy claims system into a modern digital platform across 10+ teams
  • Ran customer-insight sessions to keep the roadmap grounded in user needs
  • Helped create models and mock-ups of the digital claims flow, where the app showed different claim options depending on what the member picked, and tested the app live with real customers, including a public usability test at Frederick Mall in Kitchener; the accessibility finding was built into the app's onboarding
  • Cut paper-product costs 60% (~$1M saved) through digital transformation of core claims workflows

IDEO U

Remote · 2020 – 2025

Alumni Coach, Design Thinking

  • Coached learners and facilitated Human-Centered Design workshops
  • Translated complex concepts for non-technical audiences through weekly cohort videos and written guidance
Conversation DesignLLM Agent Behaviour DesignPrompt, Tool & Skill WritingEdge Cases, Fallbacks & HandoffsGenerative AI / LLMs (Claude)Evaluation (Judges, Golden Sets)Human-in-the-Loop ReviewUser Testing & IterationInteractive Replies (Carousels, Rich Choices)Plain-Language Knowledge ContentStakeholder ManagementRapid Prototyping (Claude Code)Presentation DesignRegulated-Environment Delivery