Sunny: Job Searching Companion I designed, built, and shipped with AI.
A side project turned real product: a warm, gamified job-search companion with an AI job scout, personalized interview prep, and Gmail automation. Designed and coded end to end in about a week of evenings using AI-assisted development, and live on the web today. It runs my own job search every day.
Impact
• Shipped a production app solo with AI-assisted development: from first commit to live in days, not months. • Imported my real 218-application spreadsheet in one click; the app now runs my daily search. • AI job scout finds fresh matching roles on a $0 infrastructure budget (free-tier Gemini with live search, plus bring-your-own-AI prompts). • Gmail automation archives "thanks for applying" noise and surfaces the emails that turn into interviews.
Deliverables
A live, installable web app built solo in roughly a week of evenings (Next.js, Supabase, and a $0-budget AI architecture), managing my real 200+ application pipeline.
Year
2026
Role
Designer, PM, and AI-assisted engineer: solo, end to end (product strategy, UX, visual design, front-end, deployment)
Context: a job search needs a companion, not another spreadsheet
After a layoff and 200+ applications, every tracker I tried felt like HR software pointed at me: spreadsheets, statuses, silence. But a job search doesn't run on data entry; it runs on momentum and morale. So I built the tool I actually needed, a companion instead of a dashboard, and opened it up to anyone job hunting.
Process: designing in the medium, with AI as the engineering team
I treated AI as my engineering team. Working in Claude Code against a Next.js and Supabase stack, I designed in the medium: ship a real feature, use it in my own search the same day, and iterate from lived friction instead of static mockups. Every product decision stayed mine; the AI just collapsed the distance between deciding and deployed. The feedback loop was hours, not sprints.
Core insight: reward effort, never outcomes
Job-search tools measure outcomes (interviews, offers), which are exactly the things a candidate can't control. That inversion is what burns people out. Three insights shaped Sunny: reward effort, never outcomes (you control applying; you don't control replies); nobody should re-type 200 applications to adopt a tracker, so imports had to be one click; and the highest-anxiety moments, rejection emails and interview scheduling, are precisely where software is usually coldest.
Design direction: one companion that grows with you
One companion character that grows with your effort: Finch and Duolingo behavioral design applied to something genuinely hard. Effort-based XP with streaks and badges, including one you can only earn by getting rejected and continuing. One-click spreadsheet import. An AI scout that hunts live postings against your profile and skips everything you've already applied to. Interview prep that assembles a phone-screen script from your own story.
Outcome: a live product running my real job search
A live, installable web app: a Today screen with funnel metrics, scheduled interviews, and AI-picked jobs to apply to; a color-coded pipeline with inline editing, bulk actions, and spreadsheet import; a prep tab with drills tailored to your background and a ready-made AI mock-interview prompt; and a Gmail sweep that files acknowledgement noise and surfaces the emails that matter. The AI runs on a deliberate $0 architecture: free-tier Gemini with live Google Search grounding, plus prefilled prompts into the Claude or ChatGPT people already pay for.
Takeaways
AI-assisted development changes what a solo designer can ship: the craft shifts from pushing pixels to making sharp product calls quickly and testing them live. Emotional UX is a real differentiator in serious software; warmth is a feature, not decoration. And building for your own problem is the fastest research loop there is: I was user zero, every single day.