A data analyst processed 1.2 million personal messages from two decades of digital history to build a personal CRM and uncover hidden patterns in his long-term social relationships.

Key Points

• The project analyzed 20 years of chat data from platforms including Telegram, Instagram, VK, and Twitter to create a structured life archive.
• LLMs were used to resolve identity ambiguities, such as nicknames, and to classify 1.2 million messages into life events, sentiment, and conversational topics.
• Findings revealed that while the author's total social network shrank by 75% over time, the annual volume of conversation-days remained consistently flat.
• Data showed that linguistic accommodation and vocabulary overlap are strong indicators of relationship health, while questioning frequency serves as a proxy for relationship bandwidth.
• The author utilized a deterministic Python pipeline and SQLite to maintain data provenance, ensuring all insights could be traced back to original source messages.

Why it Matters

This project demonstrates how individuals can leverage modern AI and data processing to gain objective insights into their personal lives and social habits. By quantifying relationship dynamics, users can identify behavioral patterns—such as a tendency to offer advice rather than support—that are otherwise invisible in day-to-day interactions.