AI agentsJobs
JobRadar-AI
An AI agent that ranks thousands of job posts against one resume, and keeps the tracker up to date itself
5,718job posts ranked against one resume
See the code on GitHub (opens in a new tab)
Proof you can check:Code on GitHub
AI agents
An agent that reads and writes
- search_jobs
- get_job
- list_applications
- get_profile
- draft_application_text
- save_job
- log_application
- update_application_status
- add_interview_note
- set_follow_up
- add_contact
One database: search results and the tracker
Lubo Bali · lubobali.com
The problem
There are thousands of job posts and the right ones hide among the wrong ones. Keyword search fails because both use the same words.
What I built
Spark pulls posts from 8 job APIs, every description becomes searchable vectors, an AI ranks each one against the resume with a reason (the best matches first), and an agent answers in plain English and logs applications itself.
Tech & why
Databricks for the heavy lifting, Lakebase (Postgres with vectors) so search results and the tracker live in one database, and Change Data Feed turns every change into analytics.
- Databricks
- Apache Spark
- Lakebase
- pgvector
- MCP
The result
5,718 posts, 60,860 vectors and an agent with 11 tools (5 that read, 6 that write), live as a Databricks App with 842 tests.
How this agent is built
Every AI agent has the same five parts. Here is what each one is in JobRadar-AI, so you can judge any agent the same way.
- ModelAn AI model through the Databricks AI Gateway ranks each job against the resume; the agent runs on Agent Bricks.
- HarnessSpark ingest, then embed, then score; the agent works through an MCP server with 11 tools.
- GuardrailsThe MCP server needs a bearer token; drafts are returned, never stored; every write goes through a named tool.
- Knowledge5,718 job posts as 60,860 vectors in Lakebase, plus the resume.
- Tools5 that read (search, get a job, list applications, profile, draft text) and 6 that write (save, log, update, note, follow-up, contact).
Similar projects
- AI agentsSkyIndex-AISearch that finds the answer even when the words don't match30 min between refreshes; only changed text is processed again
- AI agentsMarketMind AIA market analyst agent: live prices, news mood and signals, answered in plain English2.6 s average answer, up to 4 tool calls
- AI agentsSkyCast-AIA weather agent where every number in the answer comes from a tool, not the AI's memory5 tools the agent can call, each number from one of them
Have a problem like this? Get your free AI blueprint in about a minute.
