57 lines (44 loc) · 1.67 KB

CrewAI Agents - Project Decision

Current Implementation

We are NOT using CrewAI. Our agents are built-in FastAPI endpoints:

Agent	Endpoint	Function
Scout	`/api/agent/scout`	Opportunity discovery
Analyzer	`/api/agent/analyzer`	Parse & extract requirements
Matcher	`/api/agent/matcher`	Vectorized skill matching
Auditor	`/api/agent/auditor/*`	Resume evaluation + cover letter
Learner	`/api/agent/learner/feedback`	Track user behavior

Why CrewAI Was Excluded

Build issues: Requires GCC >= 8.4
Per AGENTS.md: "crewAI excluded due to build issues (requires GCC >= 8.4)"

Should We Use CrewAI?

Pros

Structured multi-agent collaboration
Built-in task delegation between agents
Easier to define agent roles/backstories

Cons

Additional dependency overhead
Requires LLM API (OpenAI/Anthropic) - adds cost
Slower (makes LLM calls for each agent)
Build issues on some systems

Current Agent Architecture

Scout Agent

Uses SerpAPI for web scraping
Falls back to database opportunities

Analyzer Agent

Regex/parsing for requirement extraction
Falls back to defaults

Matcher Agent

Uses sentence-transformers embeddings (all-MiniLM-L6-v2)
No LLM required - fast and free

Auditor Agent

Vectorized matching for resume evaluation
Gemini API for AI cover letters (with template fallback)

Learner Agent

Engagement tracking in database

Recommendation

Keep current implementation. It's:

Faster (no LLM calls for matching)
Cheaper (no API costs for skill matching)
More reliable (works without external AI services)
Easier to deploy (fewer dependencies)