A call center runs thousands of agent-customer calls a day. No human team can listen to all of them, so compliance issues and wrong bookings slip through unnoticed.
What I built
1Pulls call recordings from an SFTP server.
2Transcribes each call (Urdu, noise-tolerant).
3Runs an LLM audit scoring every call on 4 compliance parameters: customer consent, two-way communication, pricing clarity, and policy clarity (each yes / no / unclear).
4Saves structured results to a database.
5A production monitoring dashboard: KPI cards, per-parameter compliance %, trend charts, and a filterable call table with green-tick / red-cross verdicts and drill-down per call.
How it works
SFTP ingestionparallel workers, resumable and incremental.
WhisperUrdu speech-to-text for transcription.
OpenAIthe 4-parameter compliance analysis on every call.
Python + SQLitepipeline and structured results.
FastAPI + Chart.jsthe monitoring dashboard.
The result
8,000+ calls audited every day, automatically.
Cut wrong-booking complaints by ~85%.
A live dashboard so the team sees compliance at a glance and drills into any call.
See it run
Dashboard demo. All data is masked — mobile numbers hidden, no real transcripts shown.
Have calls or data you can't manually review? Let's talk.
Tell me where your team is losing time. If I can automate it, I'll show you how.