Zamesin Built a Personal CRM with Claude: 9,000 People from Telegram and 4,000 Dossiers in Six Hours
Vanya Zamesin kept forgetting to reply to people and losing referral contacts, so he built a personal CRM with contact dossiers and a task list. Claude went through an export of his private Telegram chats, identified 9,000 people, and in six hours built dossiers on the 4,000 most important ones using LinkedIn and the Apify service. A Telegram bot linked to Claude on his home computer runs the whole setup. Any founder can use this approach to turn chat chaos into a working list.
AI-processed from TG: @zamesin (Ваня Замесин); edited by Hamidun News
Anyone who communicates a lot on Telegram has hundreds of chats full of unfinished promises like "let's get on a call next week" and recommendations like "talk to this person." Memory fails, and the right contacts get lost. Boost school founder Vanya Zamesin solved this by building a personal CRM himself: a contact database where every person has a dossier and all tasks sit in one list. Claude (Anthropic's AI assistant) went through his chats, identified 9,000 people, and then, over six hours, built dossiers on the 4,000 most important ones using Apify, a service that scrapes LinkedIn. The whole system cost a $200 Claude subscription and $30 for Apify.
How it was built
The first step is almost trivial: Telegram Desktop has a feature to export the full history of your conversations, and Zamesin used it. He then set up a server and gave Claude an Apify API key so it could collect data from LinkedIn.
Next comes an important detail that is often skipped. Before processing anything, Zamesin explained his context to the model: what his goals were for entering the US market and for his Russian business. Claude read the working sessions and worked out for itself who it was dealing with. Without this, "important person" stays an abstraction; with context, the model can tell a relevant investor from a casual acquaintance.
After that, Claude was tasked with reading all the personal conversations and listing the people the author had talked to or who had been recommended to him. Zamesin did not analyze group chats, since they contain too much noise. The result was 9,000 people.
Dossiers on 4,000 people
Nobody needs all 9,000 contacts. So the author first decided what dossier he needed for each person, and then selected roughly 4,000 people to build them for. Claude, together with Apify, pulled information from open sources. The work took about six hours.
To avoid exhausting his limits, Zamesin ran the task on the lighter Haiku model at medium effort rather than the most powerful version. This is a sound practical move: bulk routine work, where the goal is to extract and consolidate facts, does not require an expensive model. The powerful model is worth reserving for complex tasks, while a lighter one is enough for a pipeline of thousands of similar operations.
Total cost: a weekly quota of the Claude subscription for $200 and $30 for Apify. The author stayed within both amounts. Ready-made CRMs with data enrichment usually charge per seat or per processed contact, whereas here the database was built on a subscription the author already had, plus a small fee for scraping.
The Telegram bot as a control panel
The interface is the most convenient part. Zamesin built a bot that forwards messages to a Claude session on his home computer, and that session has access to the CRM. He chose Telegram because it is the easiest place to forward messages, notes, and voice memos.
The CRM itself has a Today screen that gathers all tasks: who to contact and what to do. The author sends the bot new tasks or dictates what he wants: assign a task, find people, or improve the CRM itself or its logic. The result is a single to-do list in which Claude not only stores the list but also helps carry out the items.
What to keep in mind
The post is personal and contains no technical details: there is no database schema, no code, and no description of how the bot maintains its session. You cannot replicate the system exactly from it. There is also an ethical question: collecting dossiers on thousands of people from LinkedIn and private conversations runs into privacy concerns and, in many countries, personal data protection laws. In addition, everything runs on the author's home computer, so the system depends on that machine being switched on.
The post ends with an advertisement for the Boost and Boost Pro courses, but that does not diminish the recipe itself: export your data, explain the context to the model, process the array with a cheap model, and keep the expensive subscription for dialogue. If you are a founder, product manager, or salesperson with thousands of "frozen" contacts, this template can be assembled over a weekend and kept under your own control.
- Nov 21, 2026Skein Agent Erases Finished Stages From Context: 97% Success Rate, But More Expensive Than OpenCode
- Nov 21, 2026CoreWeave Owes $47 Billion on Nvidia Chips and Lost $709 Million in One Quarter
- Nov 21, 2026ClawDOS Writes and Runs Code on an IBM PC 5160: Primes Under 100 in 64 Seconds
- Nov 21, 2026A Developer Had Claude Install 900 Apps and Built a Free Mac Cleaner
Key Questions Answered
How it was built?
The first step is almost trivial: Telegram Desktop has a feature to export the full history of your conversations, and Zamesin used it. He then set up a server and gave Claude an Apify API key so it could collect data from LinkedIn. Next comes an important detail that is often skipped. Before processing anything, Zamesin explained his context to the model: what his goals were for entering the US market and for his Russian business. Claude read the working sessions and worked out for itself who it was dealing with.
What are the key details on Dossiers on 4,000 people?
Nobody needs all 9,000 contacts. So the author first decided what dossier he needed for each person, and then selected roughly 4,000 people to build them for. Claude, together with Apify, pulled information from open sources. The work took about six hours. To avoid exhausting his limits, Zamesin ran the task on the lighter Haiku model at medium effort rather than the most powerful version.
What are the key details on The Telegram bot as a control panel?
The interface is the most convenient part. Zamesin built a bot that forwards messages to a Claude session on his home computer, and that session has access to the CRM. He chose Telegram because it is the easiest place to forward messages, notes, and voice memos. The CRM itself has a Today screen that gathers all tasks: who to contact and what to do. The author sends the bot new tasks or dictates what he wants: assign a task, find people, or improve the CRM itself or its logic.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.