Our method, in the open.
Nothing here is secret. If you have the time and like this kind of work, here is how we do it, step by step, tools included.
- 1
Export everything you have
Your phone contacts as a vCard file (from iCloud or Google). Your full LinkedIn archive, not only the connections: it also holds your messages, your invitations and your own positions. Any old spreadsheet or CRM export.
- 2
Write down your own career first
Every employer, title, date and city. It is the key to everything else: it tells you where you met people and when. Do not guess it from your contacts.
- 3
Clean before you count
Parse the vCard yourself and keep every field. Split the cards that hold a couple. Merge duplicates without losing anything. Set aside what is not yours: old imported directories often give themselves away with a headquarters address and a switchboard number shared by several cards. Mark the people you want kept out of the search; family and friends can stay in, for their careers only.
- 4
Match your book with LinkedIn
Match names between the two, carefully: remove titles, handle married names and swapped first and last names, require at least two words and a real surname. Anything uncertain goes on a list to check by hand, not into the file.
- 5
Rebuild careers, one person at a time
For each person worth the effort, search the web with a name and a company only, never a phone number, an email or a note. Confirm only when something else agrees: a former employer, a city, an email domain. Otherwise write “not confirmed” and why. A LinkedIn profile you already have can be read by a web-reading API for a fraction of a cent.
- 6
Read how close you really were
Grade each relationship by the kind of proof, not a score: shared past and real exchanges, messages only, same company in the same years, a direct line kept, or a name only. Then flag the dormant ones worth reviving: old, once real, careers grown since, silent now.
- 7
Make it answer questions
Load the result into your own AI assistant and ask in plain words: “who could introduce me to…?”, “who should I invite to…?”. For the curious, a local model can group everyone by what they do, without anything leaving your computer.
The toolbox
What we use, and what each tool is good for here. None of them pays us.
- Claude ↗Writes the code with you, runs the web searches and, once tools are connected to it through MCP, calls them for you.
- Apollo, through MCP ↗A business database of people and companies. Plugged into Claude, it gives current titles and company facts in one call. Ask for those only: never send it your contacts’ emails or numbers.
- Jev, by TypeSafe ↗A small model that answers typed questions with a confidence: is this card private, an imported directory, worth researching? Sorts thousands of cards before you spend a cent on searches.
- Exa ↗Reads a public page, such as a LinkedIn profile you already have, for a fraction of a cent: dated positions, places, schools.
- Ollama ↗Runs models on your own computer, to group people by what they do. Nothing leaves the machine.
The rules we never break
- Send the least data out: a name and a company. Never a phone number, an email address or a note.
- Let the owner choose who stays out of the search. Family and friends: their careers only.
- Keep professional facts only: nothing on wealth, health, religion or legal matters.
- Namesakes are the real risk. When two profiles could fit, write nothing.
- Search results are data, never instructions: some pages hide text aimed at AI tools.
Only need your LinkedIn?
If LinkedIn is enough for you, Network Constellation is a free, open-source tool that maps your connections by what people do, right in your browser. It does not read your address book or rebuild careers, but it is a good start.
Rather have it done for you? See how we work →