AGNT-201

Inspiration

Job boards are broken. 90% scam listings, zero feedback loops, endless noise. We're engineers who've been through this hell. So we built what we wished existed: an AI agent that hunts jobs like a obsessive friend who actually gets what you're looking for.

What it does

Upload your resume, get back ranked jobs with hard metrics:

1. Alignment % (embedding similarity + weighted keywords)
2. Fakeness score (post age, duplicate detection, scam patterns)
3. Full reasoning trace for every match

No scrolling. No guessing. Just signal.

How we built it

1. Frontend: AI chat interface (Next.js, aisdk)
2. Backend: Next.js, Python, async worker pipeline for scraping + enrichment
3. ML: SentenceTransformer (all-MiniLM-L6-v2) embeddings for semantic matching; selective use of LLMs for explanations
4. Data: Lean schema designed for speed, not complexity
5. Scraping: Multi-source (LinkedIn, Wellfound, YC Jobs)

Challenges we ran into

1. Job platforms actively fight automation
2. Resisting feature creep – we killed good ideas ruthlessly

Accomplishments that we're proud of

1. Built real defensible signal around job quality
2. Created something that feels like an agent, not a demo

What we learned

1. LLMs are scalpels, not hammers – use sparingly
2. The job search problem is deeper than UX – it's incentive misalignment

What's next for AGNT-201

1. Interview predictor at full scale
2. Agent memory for personalized job alerts
3. Aggressive deduplication across all major platforms
4. Ship to real users and iterate based on actual job search outcomes

Built With

• aisdk
• jupyter
• nextjs
• python