Anti-Soy

Startups are struggling to find talent and talent don't hear about startups.We enable startups to find talent judging by the talents' real world coding skills.

Comment 1

https://ericjujianzou.github.io/Anti-Soy/

Inspirations

As students, we find the job process difficult, almost impossible if we don't have a network or referral.

We verified our idea by doing user interviews with sponsors at Nexhacks from Kairo, Arize, and LiveKit about their hiring process. We learned that startups often spend a lot of time and capital on acquiring talent, because many talents haven't heard about them before. Chris from LiveKit told us that they've done multiple hires from the LiveKit community, and LinkedIn outreach, which proved to be more effective than resumes.

So we came up with the idea of judging applicants' skill through their real world projects.

Then we realized 2 startups are already doing what we're doing. One of them, called SkillSync, is YC-backed.

We tested out their projects, and noticed a crucial flaw - their apps only do a surface level scan on tech stack and general info. This offers little value because any soy dev (The Silicon Valley/ Seattle based soy latte drinking hipster you see at the coffee shop, coding NodeJS on his macbook pro) can use AI to make their github project look "professional"

For example, I put my profile in SkillSync and it recognized my EMPTY repo as "Implements Model Context Protocol servers for managing context across distributed AI agent systems"

This gave us confidence that our problem is real, and that we can do better than our competitors.

What it does

Our product detects if a repo has a lot of AI usage, and if it follows production-level standards. We have 18 metrics that determine this, and use LLM-judge and statistical (no AI! Surprise!) methods to perform calculations on the two fields. Some companies might actually want good prompt engineers, so AI != bad!!!

How we built it

We actually didn't write a single line of code until 9 pm Saturday, meaning 8 hours of walking around talking to people, brainstorming, debating, and designing.

Our stack is actually really simple! ( ANTI SOY! ) We have a python + FastAPI backend, Next.js frontend, and Sqlite DB.

Less than 2000 lines of code for backend logic -> put/fetch from DB -> display on frontend.

P.S. We believe we win technically complex track award because the hardest thing for devs is to write complex logic using simple code. ANTI SOY!

Challenges we ran into

  • Taking the risk to challenge a YC startup on the same problem.
  • Planning which metrics to evaluate from a codebase
  • Hitting Gemini quotas because we had too many metrics => too much calls.
  • Argument between Eric and Andrew because Eric used too much Copilot (ANTI-SOY!)

Accomplishments that we're proud of

  • Using relatively small code base and development time to flesh out fully functional MVP
  • Spending almost as much time for problem discovery as development

What we learned

  • AI coding DOES complicate a lot of things unnecessarily (E.g. generating a folder that just contains 1 file to do a particular thing)
  • Designing first speeds up development A LOT
  • Competition can mean you are on the right path

Learnings from using DevSwarm and Trae:

  • Shipping exposes real problems faster than planning
  • We can build really fast if we already have good knowledge
  • Parallel thinking beats sequential execution

What's next for Anti-Soy

  • GitHub Search API (find devs by language/location)
  • GitHub Trending scraper (identify rising developers)
  • Community parsing (HackerNews, Reddit, Dev.to mentions)

Get our first customers - Please reach out if you have intros!!! @ [email protected] Also refining our algorithms to build a moat so the competitors can't copy. Since they're startups they will copy fast.

Pricing model

Startups pay us a % upon securing a hire.

Built With

Share this project:

Updates