Building in Public: Our First Month
A month ago, Wrok was a sketch in a notebook and a half-finished prototype. Today it's a working product with real users. Here's what that month looked like.
The stack we chose (and why)
We wanted to move fast without sacrificing the ability to iterate. The stack:
- Next.js for the frontend — App Router, server components, streaming
- Python + FastAPI for the backend agent — LLM orchestration, database operations
- Supabase for auth and Postgres — instant setup, good DX
- Fly.io for deployment — simple, close to users, Docker-native
The key decision was separating the AI agent into its own service. It would have been faster to put everything in Next.js API routes, but we knew the agent would need its own scaling characteristics and deployment cadence.
What worked
Voice input. We added it almost as an afterthought — let people talk about their work instead of typing. It turned out to be the feature people loved most. Talking about your career is more natural than writing about it, and the transcription quality is good enough that most people don't need to edit much.
Conversational onboarding. Instead of a form with 20 fields, we ask one question: "Tell us about your work." Then the AI agent asks follow-up questions, extracts structured data, and builds a profile. This felt risky — what if people don't know what to say? — but it worked better than expected.
What didn't work
Too many landing page iterations. We went through six versions of the landing page before we had users. That's five too many. We should have shipped the first one that was "good enough" and iterated based on real feedback.
Overengineering the PDF service. We spent days building a flexible template system for PDF generation when all we needed was one clean template. The flexibility wasn't useful until we actually had multiple use cases, which we didn't.
What surprised us
The thing that surprised us most: people don't want to just generate a resume. They want to understand their own career better. The AI conversations often turned into genuine reflection — people discovering patterns in their work they hadn't noticed before.
That insight is shaping everything we build next.
What's next
We're focused on three things:
- Better narrative extraction — helping people find the story in their work experience
- Profile pages — public profiles that showcase your career story, not just your job titles
- Integrations — pulling in context from GitHub, LinkedIn, and other places where your work lives
More updates coming. Follow along.