Jobs Radar

Personalised job search pipeline: scrape → tag → embed → rank → rate.

Features

• Resume parsing — paste raw resume text, get structured output (target roles, skills, seniority, experience) via OpenAI structured outputs
• Parallel scraping — scrapes Indeed across multiple queries and locations concurrently using ThreadPoolExecutor
• Dedup + tagging — removes duplicate listings, tags each job for seniority, entry-level, remote, clearance, and experience range using regex
• Hybrid vector search — dense (BGE) + sparse (BM25) search with Reciprocal Rank Fusion via Qdrant in-memory
• LLM-as-judge rating — each result is scored 1–10 against the candidate's resume using gpt-4o-mini
• Feedback loop — rate jobs 1–5 stars; liked/disliked examples are injected into the judge prompt on future searches to personalise ratings over time
• Single-table persistence — search results and user feedback live in one SQLite table (job_results), keyed by job_url; re-submitting feedback updates the row, no duplicates
• CSV export — every scrape is auto-saved to ~/.jobs-radar/jobs_{timestamp}.csv, downloadable from the UI
• Streamlit UI — parse resume, search jobs with inline feedback, browse and rate recent results
• REST API — FastAPI backend, useful for scripting or other clients

Stack

• Python 3.12, FastAPI, Streamlit
• OpenAI API (gpt-4o-mini by default)
• Qdrant (in-memory) + fastembed
• jobspy (Indeed scraping)
• SQLite (results + feedback storage)
• uv (dependency management)

Installation

pip install jobs-radar
# or with uv
uv add jobs-radar

Or run from source:

git clone https://github.com/your-username/jobs-radar
cd jobs-radar
uv sync

Setup

cp .env.example .env  # fill in OPENAI_API_KEY

Data (SQLite DB + CSVs) is stored in ~/.jobs-radar/ by default. Override with DATA_DIR in your .env.

Running

You need two terminals — one for the API server and one for the UI.

From a pip/uv install:

# terminal 1 — start the API server
jobs-radar-serve

# terminal 2 — open the UI
jobs-radar-ui

From source:

# terminal 1 — API
uvicorn jobs_radar.main:app --reload

# terminal 2 — UI
streamlit run jobs_radar/ui.py

Then open http://localhost:8501 in your browser. API docs at http://localhost:8000/docs.

Configuration

All settings are read from environment variables or a .env file:

API Endpoints

Project Structure

jobs_radar/
  main.py          # FastAPI routes
  pipeline.py      # scrape, dedup, tag, validate
  vector_store.py  # Qdrant setup, upsert, hybrid search
  llm.py           # resume parsing, job rating, judge prompt
  feedback.py      # SQLite — job_results table, save/load functions
  models.py        # Pydantic models
  config.py        # pydantic-settings — all env/config vars
  cli.py           # jobs-radar-serve and jobs-radar-ui entrypoints
  ui.py            # Streamlit UI
tests/             # pytest suite (54 tests)

Data Model

All results and feedback share one SQLite table (~/.jobs-radar/feedback.db):

Re-running a search updates system fields for any URL already in the DB while preserving user_rating and notes.

Notes

• Qdrant runs in-memory — the vector index is not persisted between server restarts. The SQLite DB and CSV exports are the durable layer.
• fastembed downloads model weights on the first upsert_jobs call (~30–60s). Subsequent calls are fast.