M-Pesa Insight Agent
M-Pesa is Kenya’s most widely used mobile money platform, generating statements filled with PayBill payments, Till purchases, transfers, airtime purchases, withdrawals, deposits, subscriptions, and recurring transactions. While the data exists, most users struggle to analyze long PDF or CSV statements to answer simple questions like:
“What am I spending most on?”, “Who receives most of my money?”, or “How much goes to bills monthly?”
I built an AI-powered financial insight agent that transforms raw M-Pesa statements into a context-rich knowledge graph using Neo4j Aura.
The agent allows users to query their statements in natural language:
-
“How much did I spend on bills last month?”
-
“Who do I send money to most?”
-
“Show my highest spending categories.”
-
“Which transactions repeat frequently?”
Using Neo4j graphs, the agent connects users, merchants, transaction types, categories, and spending patterns to uncover meaningful financial insights instead of just displaying raw transactions.
Why Graphs?
Financial data is naturally connected. Every transaction links people, businesses, categories, dates, and behaviors. Neo4j makes it possible to reason over these relationships and identify patterns traditional tables cannot easily reveal.
Tools Used
-
Text2Cypher for converting natural language into Cypher queries
-
Vector Search for semantic understanding of transaction descriptions and merchant similarity
-
Neo4j Aura Knowledge Graph
This project turns unstructured M-Pesa statements into conversational financial intelligence, helping users understand their money without manually searching through PDFs or spreadsheets.
