Context
Hiring great designers is one of the highest-leverage things I do as a design leader — but also one of the most time-consuming. Across hiring loops I noticed the same pattern: I was asking the same questions repeatedly, candidates were asking the same questions repeatedly, and interviews often spent the first 15–20 minutes establishing context before deeper conversations could begin. At the same time, I wanted to build hands-on experience with modern AI systems and tools like Figma Make.
Could I redesign the hiring conversation itself using AI?
Insight
Instead of candidates interviewing me only after scheduling a call — what if they could "interview me" anytime? What if candidates could explore how I think about design, how I evaluate product decisions, how I lead teams, and how Step builds products — before we ever spoke live? This led to a simple hypothesis: if I could build an AI trained on my work, experience, and thinking, it could act as a scalable version of me in early hiring conversations.
Hypothesis
- Answer common candidate questions instantly
- Give candidates deeper insight into how I think and work
- Reduce repetitive interview time
- Improve candidate alignment before interviews begin
- Turn hiring conversations from information transfer into meaningful discussion
Solution
I built a custom GPT-powered chatbot trained on my resume, design case studies, leadership philosophy, past writing, and product decision frameworks. The bot allows candidates to ask questions like: How do you evaluate designers? What was the hardest design problem you solved at Step? What do you value in product designers? Instead of static portfolio pages, candidates can have a conversation with my experience.
Process
Phase 1 — Rapid Exploration
I intentionally started scrappy — using basic tooling and LLM APIs to prototype quickly and test whether the idea was interesting before investing heavily in architecture. The goal was speed of learning, not technical perfection.
Phase 2 — Early Testing
I shared the prototype with designers, peers, potential candidates, and recruiters — observing what questions people asked, where the AI failed, and where responses felt surprisingly valuable. One thing became clear immediately: the quality of answers depended far more on structured inputs than on the model itself.
Phase 3 — Building the Knowledge System
I built a knowledge base including structured case studies, resume data, team documentation, leadership philosophy, and product decision frameworks. This evolved into a lightweight RAG-style system where the AI could reference relevant context dynamically.
Phase 4 — Prompt & Response Design
Prompt design became the core product layer. I iterated on structured prompts, response tone and voice, contextual grounding, and hallucination reduction — frequently using AI itself to optimize prompts and knowledge formatting. The UI remained intentionally minimal, prioritizing conversation over interface complexity.
Key Design Decisions
- Start messy on purpose — optimize for learning speed rather than technical correctness
- Treat prompts as product — prompt architecture is the primary UX layer
- Invest in knowledge design — structuring experience into usable data unlocked the biggest improvements
- Keep the cost of failure low — the project restarted multiple times with minimal friction
Impact
- Reduced repetitive interview questions
- Candidates entered conversations with stronger context
- Interviews shifted toward higher-level discussion
- Gained hands-on experience building AI systems
Why This Matters
For design leaders, time is the most limited resource. AI systems like this allow leaders to scale knowledge, reduce repetitive communication, improve hiring quality, and focus conversations on real signal instead of context building. In many ways, this project was less about chatbots and more about a new question.
What happens when your experience becomes a product?
Future Exploration
- Integrating the bot into hiring funnels
- Tailoring responses based on role or seniority
- Expanding the system into mentorship and design education
- Applying similar systems to internal design knowledge sharing