Two lanes of discovery
Seven public aggregators, plus an 884-company registry resolved into company job boards and your own watchlist.
AI-assisted workflow · Web · Local by design
employed
Find the fit.
Keep the evidence.
A local workspace that finds remote roles a person could plausibly win, explains how to approach each one, and tracks what they did about it.
- Role
- Design and full build
- Platform
- Local Python web app
- Data
- Stays on your machine
Core features.
From thousands of postings to the few worth a serious application.
A filter cascade
Geo gate, seniority gate, anti-terms, then an unambiguous title signal, before anything is scored.
Analysis on demand
Each card has its own Analyse button. The agent classifies eligibility, quotes the posting, and builds a requirement-to-evidence matrix.
A profile built from documents
CVs, portfolios and notes are read into proposed capabilities, each tied to the exact span of the document that proves it.
One board from arrival to outcome
Ranked, Analysed, Applied, Interview, Lost. No drag and drop and no JavaScript library.
Bring your own model
OpenAI-compatible, Anthropic and Google formats, local models through Ollama or LM Studio, or the coding agent itself.
Architecture
From thousands of postings to evidence you can trust.
Hover or tap a component to trace its connections.
Job sources
Seven aggregators plus resolved company boards
- JSON
- RSS
- ATS APIs
- Harvester · postings
Harvester
Fetches and normalises postings politely
- httpx
- selectolax
- Filter cascade · normalised
Filter cascade
Geo, seniority, anti-terms and title signal, then ranking
- Deterministic
- Analysis agent · shortlist
Evidence check
Rejects any quote that is not verbatim in its source
- Pydantic
- Local store · verified only
ATS resolver
Detects each careers page's job board and verifies it live
- Greenhouse
- Lever
- Ashby
- Job sources · verified boards
Profile documents
CVs and notes read into evidenced capabilities
- pypdf
- python-docx
- Analysis agent · evidence spans
Analysis agent
Eligibility, requirement matrix and approach notes
- Any provider
- Degraded mode
- Evidence check · verdicts + quotes
Local store
Postings, profile, runs and decisions in ~/.employed
- SQLite
- Dashboard · review and decide
Command line
Harvest, resolve, match and analyse on a schedule
- Typer
- Rich
- Local store · runs
Dashboard
Board, job workspace, profile, sources and agent pages
- FastAPI
- Jinja
- HTMX
All connections
- ATS resolver to Job sources: verified boards
- Job sources to Harvester: postings
- Harvester to Filter cascade: normalised
- Filter cascade to Analysis agent: shortlist
- Profile documents to Analysis agent: evidence spans
- Analysis agent to Evidence check: verdicts + quotes
- Evidence check to Local store: verified only
- Command line to Local store: runs
- Local store to Dashboard: review and decide
Innovation highlights.
Innovation one
Evidence the model cannot invent
Every ingested answer is checked: the quote must be a verbatim substring of the posting or document. A hallucinated quote is refused before it reaches the database.
Innovation two
Degraded mode is a supported state
Without an agent, harvest, the geo gate, matching and ranking still run. The dashboard names what is dark and why, rather than faking it.
Innovation three
A board token is a hypothesis
A company's job board only counts once its API answers with a 200 and at least one real posting. Nothing becomes a source on a guess, and roughly a third of the registry resolves.
Technology stack
What it is built with.
Application
- Python
- FastAPI
- Uvicorn
- Pydantic
Interface
- Jinja
- HTMX
- No build step
Harvesting
- httpx (HTTP/2)
- selectolax
- Playwright (optional)
Data and documents
- SQLite
- pypdf
- python-docx
- PyYAML
Tooling
- Typer
- Rich
- pytest
Next project