LangChain

Customer calls turned into answers anyone can look up

Turned LangChain's customer calls into answers anyone can look up, like which customers run on which cloud or need self-hosting

Vishnu Suresh Software Engineer, LangChain LinkedIn

The problem

Once LangChain's GTM agent became a Slack chatbot, people across sales, product and engineering started asking it questions like "which of our customers are on Azure?" or "what LLM provider does this customer use?" The answers weren't in the CRM or the data warehouse, so the agent couldn't answer them well.
What they built

A nightly job that reads every customer call transcript and pulls out about 50 fields the team agreed on: whether the customer is building agents for internal or external use, which cloud they build on, who the champions and stakeholders are, and security needs like self-hosting or data residency. A weekly job tracks how each customer's answers change over time, so the agent can explain what has happened with a customer, for example one at risk of churning.
So what

What customers say on calls stops living only in recordings and reps' heads. Anyone can ask which accounts share a requirement, or what changed with a customer lately.

Under the hood

Call recordings are transcribed and land in BigQuery. A nightly cron runs each new transcript through an LLM extraction pipeline for about 50 signals and stores the results in BigQuery. A job compares extractions week over week to build a per-customer state or contact graph over time. The agent, available as a Slack chatbot and used by one rep as an MCP from Claude Code, answers questions by writing SQL against the warehouse, which Vishnu found worked better than going through APIs.