Back to work
Customer Context Engine
pocZendesk Data Platform, 2025
Turns unstructured customer signals (Gong call transcripts, UserVoice feedback) into a queryable, ontology-classified layer on Snowflake and Neo4j.
SnowflakeNeo4jLLM classificationGongUserVoicePython
Problem
The richest customer signal in the company sat in call transcripts and feedback threads nobody could query. Product and sales teams were reading transcripts by hand to answer 'what are customers asking for?'
Approach
- Layered architecture: raw transcripts and feedback land in Snowflake; an LLM pass classifies each into a governed ontology of themes, products, and sentiment; relationships between accounts, themes, and requests are modelled as a graph in Neo4j.
- Ontology is versioned so re-classification is reproducible when definitions change.
Impact
- Structured, aggregatable insight from thousands of unstructured customer conversations.
- Proof of concept validated with product and go-to-market stakeholders.