Compass: From a Friend’s Request to a Working App

Compass started with a request from a friend. I built it into a working app for finding candidates on LinkedIn, reviewing the evidence behind a match and keeping that research together for later.

Compass search criteria with a role description, Anaplan skills, credentials, locations and minimum experience — Review the role description, skills, credentials, locations and minimum experience before continuing to search.

The Problem

Finding a profile is only one part of candidate research. The next questions are harder: which requirements does this person appear to meet, what evidence supports that assessment and how do they compare with someone else? A useful shortlist needs more than names and links.

Compass brings those steps into one workflow. The role criteria, downloaded profiles, review notes and comparisons stay connected, so a search can be revisited without starting the research again.

From Role Brief to Shortlist

Define the role. Enter skills, credentials, locations and minimum experience, with optional preferences. Criteria are entered explicitly; a pasted job description is not automatically interpreted.
Run a search. Compass retrieves LinkedIn profiles through the signed-in connector, downloads new profiles and scores them using the configured criteria. Existing downloads are reused.
Review the list. Check candidate names and duplicate sources, add a review note and confirm the list before opening the ranking.
Inspect and compare. Review the saved evidence for an individual candidate, or compare two or three people side by side.
Return later. Saved searches keep previous runs together, with profiles and evidence available locally even when the connector is offline.

Make the Score Explainable

A score is useful only when the person reviewing it can understand what sits behind it. Compass lets the reviewer open a candidate’s score breakdown and inspect the saved evidence, rather than treating the ranking as the end of the process.

Compass candidate review drawer with a fictional candidate’s score and signal breakdown — The candidate review view exposes the score and supporting signals for closer inspection. The candidate shown is fictional.

Scoring weights can be adjusted in Settings, which recalculates saved evidence locally. Connection distance controls search reach but does not affect scores. Confidence describes evidence availability; it is separate from how closely the retrieved evidence matches the role criteria.

Built to Run Locally

Searches, profiles and evidence are stored in a local SQLite database. LinkedIn retrieval uses the user’s signed-in session, while saved work remains available for review offline. Compass does not send messages or connection requests.

The app starts with a single command, ./compass, on macOS or desktop Linux. The launcher prepares the dependencies, builds the frontend when needed, starts the local services and opens the app. The first launch includes an interactive LinkedIn sign-in; later launches reuse the installation and login.

Python backend coordinates the application and its local services.
Vite frontend provides the role brief, results, candidate review, comparisons, settings and saved-search views.
SQLite storage keeps searches and retrieved evidence on the user’s computer.
LinkedIn MCP connector provides LinkedIn access through the open-source linkedin-mcp-server project by Daniel Sticker and its contributors.

The Result

A friend’s request became a complete path from role criteria to candidate comparison. Compass brings search, evidence review and saved research into the same app, with an interactive guide to help someone work through the process.

The part I want to highlight is the connection between those steps. Finding candidates, understanding a score and returning to an earlier search all belong to the same piece of work. Building Compass meant making that whole path usable.

Started with a friend’s request. Built into a tool for a real workflow.