In this episode hosts Philip Maymin and Jie Tao, DSc explore the next evolution of Health Buddy, an AI tool built to help users navigate complex medical information safely and reliably. Jie explains how version 2 introduces a self-healing architecture. One that can recover from failure, rebuild workflows, and adapt through “editor-in-chief” and “planner” agents working together. He discusses what it takes to balance autonomy, ethics, and control, and how a thoughtful “trust layer” protects users from harmful or misleading content. Listeners gain insight into the practical difference between simple workflows and fully agentic systems and why reliability and resilience matter most.

Learn more about Fairfield Dolan's AI and Technology Institute.

In this episode Jie Tao, DSc shares the story behind building Health Buddy, an AI-powered assistant designed to help people make sense of complex medical information. What began as a personal late-night search for answers evolved into a sophisticated multi-agent system built entirely by one developer with the help of AI. Jie breaks down the design process, from automating research and review to introducing “human-in-the-loop” feedback and testing. He reflects on the challenges of hallucination, reliability, and the moral weight of creating tools that could impact real lives. This conversation offers a candid look at how curiosity, iteration, and ethics drive responsible AI innovation.

Learn more about Fairfield Dolan's AI and Technology Institute.

Go from being an AI user to an AI architect. This episode provides the blueprint for automation, teaching you how to use "Context Engineering" to make your AI a specialist and when to choose a simple workflow versus a powerful agentic system.

Learn more about Fairfield Dolan's AI and Technology Institute.

Chapters: 00:00 – Welcome back & episode setup 01:05 – Reframing automation: elevate the goal, not just the speed 02:40 – Audit the workflow: steps, transitions, and quick wins 05:10 – Crawl–walk–run: introducing change people will adopt 08:30 – Picking the first high-reward step for AI 10:55 – From one big prompt to linked tasks and roles 12:20 – What is an agent? Tools, context, and autonomy 15:20 – Quality control with examples (few-shot learning) 16:35 – Choosing examples without constraining creativity 19:50 – Turning examples into reusable prompts (meta-prompting) 21:36 – Bringing in private knowledge with RAG 25:10 – Tools and setup for RAG and context management 33:13 – Common pitfalls and how to build reliable RAG systems 41:00 – When to stop prompting and start orchestrating 48:41 – Using multiple tools and managing credibility 51:26 – Scaling responsibly and keeping humans in the loop 53:51 – Downloadable blueprint and automation roadmap 55:37 – Lessons learned and preview of next episode

Hosts Philip Maymin and Dr. Jie Tao, from Fairfield Dolan and the University’s new AI & Tech Institute, show how to move beyond “prompt and pray” toward reliable, structured results with AI. Hear why most one-shot chats fall short, how to “flip the switch” so AI asks clarifying questions, and why treating AI like a brilliant, but clueless, intern leads to better outcomes. You’ll learn the four-part “AI job brief” approach (including role and audience), when to iterate, and how to make structured prompts statistically inevitable—rather than hopeful guesses.

Learn more about Fairfield Dolan's AI and Technology Institute.

Chapters: 00:00 – Welcome, hosts, and episode focus 01:33 – A perfect brief vs. a painful miss: instructions matter 02:00 – What “prompt and pray” is—and why it fails 03:09 – One-shot chats, search habits, and better use of AI 05:17 – The generalist-doctor analogy: give problem context 08:36 – Flip the Q&A pattern: make AI ask follow-ups 15:15 – Iterate your prompt like drafts (not one and done) 16:15 – The backspace rule: why iteration beats perfection 17:01 – Structured prompts part 1: define the role 19:30 – Structured prompts part 2: set the audience 26:59 – Memory windows: keep briefs lean and focused 34:53 – Working in steps: sequencing tasks with RAFT 41:56 – From “pray” to “statistically inevitable” results