How it works

1. A daily job pulls about 1,900 Summer 2027 internships from the SimplifyJobs list.
2. Where a posting is on Greenhouse, it also pulls the full job description.
3. Each posting is tagged with skills and turned into a vector.
4. When you upload a resume, the API turns it into a vector too.
5. It finds the closest postings and scores them:
   fit = 0.6 × meaning match + 0.4 × skill overlap
6. You get the top 20 jobs, the skills you are missing, and courses that teach them.

Stack

• Website: Next.js
• API: FastAPI (Python)
• Vectors: fastembed (bge-small, 384 numbers per text)
• Database: Supabase Postgres with pgvector
• Daily refresh: GitHub Actions
• Hosting: Render

Privacy

• Your resume is read in memory and then deleted.
• It is never saved or logged.

Plan

• Part 1 (Oct 13 to 18): job finder
• Part 2 (Oct 19 to 25): course finder with prerequisite chains

How I measure it

• Job matches: 10 test resumes, 5 good-fit jobs each. Goal: 3 of 5 in the top 10.
• Course search: 30 test questions. Goal: the hybrid search beats keyword and vector search alone.
• Speed: under 1.5 seconds per match.

What I am learning

• Vector search and ranking
• Measuring search quality with test sets
• Building a data pipeline that runs every day