AI Career Coach

Problem

Members lacked confidence and direction. Traditional job platforms didn't feel like they were in their corner.

Members often struggled with confidence and direction in their job search. Many didn't know how to start, lacked guidance, or didn't feel represented by traditional job platforms. Leadership needed a solution that felt personal, supportive, and approachable. Fully lifelike talking avatars weren't financially feasible at scale.

Hypothesis

"If members could choose a coach that felt relatable and alive, they would feel more engaged and supported, leading to higher application rates and stronger retention."

Research

Photorealistic coaches won. Job searching is serious.

A preference test with 30 participants compared illustrated and photorealistic coach styles to determine which felt most appropriate for a job search context.

• 22 of 30 participants preferred photorealistic coaches over illustrated ones
• Illustrated styles were described as "cartoony, silly, and game-like"
• Photorealistic coaches were described as "professional, trustworthy, and relatable"
• Job searching is serious. Realistic coaches matched member expectations.

Coach Development

18 curated personas built with MidJourney.

Using MidJourney, I explored a wide range of coaching personas varying in age, attire, tone, and cultural background. The goal was a library that felt genuinely diverse and representative, not a token collection. After extensive iteration, 18 coaches emerged, each with a distinct persona that felt approachable and trustworthy.

Animation & Motion

Static portraits felt impersonal. Runway brought them to life.

I experimented with Runway's Act-1, Gen-3, and Gen-4 models to add motion to the coach portraits. Finding the right prompting approach took significant iteration.

• Short, simple prompts ("slight head tilt, natural smile") produced the most lifelike results
• Long descriptive prompts created uncanny distortions and artifacts
• Idle animations like blinking, head tilts, and subtle smiles proved more scalable than full lip-sync
• Gen-3 and Gen-4 models outperformed Act-1 for natural motion quality

Short, simple prompt with natural idle motion

Long descriptive prompt with uncanny distortions

Onboarding Flow

A four-step customization sequence that made the coach feel truly theirs.

Chat Experience

Designing conversations that guide, not just respond.

Once onboarding was complete, the experience shifted into an ongoing relationship between the member and their coach. The challenge was designing a chat experience that felt helpful, trustworthy, and action-oriented, rather than reactive or generic.

01

Proactive Guidance

Designed the coach as a guide, proactively suggesting next steps like improving a resume or exploring relevant jobs to reduce decision fatigue.

02

Actionable Momentum

Structured conversations into small, actionable steps with concise responses, reinforcing progress through follow-ups and subtle encouragement.

03

Real-World Persistence

Designed for real-world usage patterns with persistent context and proactive re-engagement. When members returned, the coach picked up naturally, surfacing timely prompts like searching for new jobs or following up on recent activity to keep momentum going.

Results