Skip to content
← Projects
AI Observability
2025–Present

Realtime Analytics Watcher

A real-time monitoring dashboard for conversational-AI performance — retention, latency, and token cost tracked like a live systems watcher, not a monthly report.

PythonMLflowFigmaData Visualization
01

Problem

Once conversational agents are running in production across multiple brands, retention, latency, and cost can each degrade silently — without a real-time view, issues only surface once someone complains.

02

Approach

Built live dashboards surfacing the KPIs that actually predict a bad customer experience: retention and abandonment rate, session volume by channel, latency broken down by use case and FAQ, and token consumption per hour.

03

Architecture

Session and interaction logs → metrics pipeline → real-time dashboard (KPIs, latency breakdowns, token/cost tracking) → visual flags on critical thresholds like abandonment rate.

04

Results

  • Deployed as the monitoring layer for conversational agents across multiple retail brands.
  • Surfaced latency and cost patterns by use case and FAQ, guiding where to optimize first.
05

Future work

Adding automatic alerting when KPIs cross critical thresholds, instead of relying on someone checking the dashboard.