Sentiment AI analyses university call centre conversations. I joined as a founding engineer and built much of its AI infrastructure.
Multi-agent batch processing
Designed and built a parallel LLM pipeline that reads call transcripts in batch and scores them on behaviour, compliance and causes, using Azure OpenAI, Claude and Gemini. 8 unit tests and 6 integration tests.
Sales automation
Built a full CRM and LinkedIn automation system from scratch. It classifies leads, sends connection requests, schedules follow-ups and enriches leads from Apollo.io and LinkedIn Sales Navigator. Account-based prompt pipelines write the C-suite outreach.
Platform and infrastructure
Contributed to the production Flask REST API (JWT auth, Redis caching, deployed on AWS ECS), Next.js coaching dashboards, and Apache Airflow DAGs for transcription, speaker assignment and 5 parallel LLM analyses.
Reporting
Built Apache Superset dashboards for executives, covering interaction volumes, enrolment rates and agent performance.