Job Seeker Agents
The Problem
Job searching in AI/ML is a time-consuming, repetitive process that doesn't scale:
- Browsing dozens of job portals daily
- Manually evaluating each offer against your skills
- Customizing CV for every single application
- Losing track of applications and follow-ups
- Wasting hours on positions that don't match
The real pain: Available tools either don't personalize CVs properly, or they're complete black boxes with no control over the output.
My Solution
I built Job Seeker Agents — a multi-agent AI system that automates 80% of the job search process while keeping the human in control.
Key innovation: Instead of a single monolithic AI, I designed 4 specialized agents that work together:
| Agent | Purpose |
|---|---|
| Triage Agent | Evaluates job fit (0-100 score) in seconds |
| CV Tailor Agent | Personalizes CV for each offer and company |
| Validation Agent | Ensures zero hallucinations in generated content |
| Review Agent | Prepares interview notes and talking points |
How It Works
Job Offers → Triage (fit score) → Trello Board → CV Generation → PDF + Interview Prep
↑
Human approval here
Human-in-the-loop design: Every offer lands on a Trello board where I can review scores, move cards, and approve which ones get personalized CVs. Full transparency, full control.
Technical Highlights
Semantic Project Matching
Instead of keyword matching, I use embeddings to find which of my projects best match each job. "FastAPI" matches "REST API", "ML" matches "Machine Learning" — context matters.
No Hallucinations Policy
The Validation Agent cross-checks every generated CV against my master profile. Skills, dates, project names — everything must exist in the source of truth.
Smart Rate Limiting
Pipeline runs in phases with batch processing to stay within OpenAI API limits while processing dozens of offers efficiently.
What I Built
- Multi-agent orchestration with OpenAI Agents SDK
- Semantic search using text-embedding-3-small
- Trello integration for workflow management
- React frontend for CV preview and configuration
- PDF generation with React PDF Renderer
- Full observability with Langfuse tracing
Skills
| Category | Technologies |
|---|---|
| AI/LLM | OpenAI GPT-4o, Agents SDK, Embeddings |
| Backend | Python, FastAPI, Pydantic, SQLite |
| Frontend | React, TypeScript, Tailwind CSS |
| Integrations | Trello API, Obsidian |
| DevOps | Docker, Langfuse, GitHub Actions |
Results
- ✅ 80% automation of job search workflow
- ✅ < 30 seconds to generate personalized CV
- ✅ Zero hallucinations — validated against source data
- ✅ Full traceability — every AI decision is logged
- ✅ Human control — approve every step via Trello