root@mindgraph:~# cat case-studies/ai-program-management-platform.md
AI Program Management Platform
DOCUMENT: AI Program Management Platform
INDUSTRY: EdTech
PROJECT: Live Training Operations
CLASSIFICATION: CASE STUDY // PUBLIC RELEASE
SUMMARY:
A platform that automatically scores every live training session and manages instructor availability before a cancellation becomes a crisis.
DETAIL:
Multi-modal session scoring (audio, video, chat) across nine benchmarks, weekly PM reports with timestamped clips, and a WhatsApp availability bot that auto-queries replacements and hands the PM a one-tap confirmation list.
IMPACT:
PM TIME SAVED: 72 hours/month of session review
20 / EDTECH / LIVE TRAINING OPERATIONS
A platform that automatically scores every live training session and manages instructor availability before a cancellation becomes a crisis.
Client: Live-training delivery company
Tags: EdTech · Live Training · Video Intelligence · WhatsApp AI · Program Management
// Problem
Program managers reviewed every trainer's 3–5 hour live sessions manually to catch quality problems — disengagement, poor interaction, late starts — that often went unnoticed until real damage was done. A senior PM was spending roughly 72 hours a month just watching sessions, and last-minute instructor cancellations were often flagged too late to arrange a replacement.
// Our Solution
A platform that automatically scores every live training session and manages instructor availability before a cancellation becomes a crisis. Every session's audio, video, and chat is evaluated against nine quality benchmarks — camera-on time, presentability, student interaction, engagement, punctuality, connection quality, chat sentiment, and language — and rolled into a weekly per-trainer report the PM can review in minutes instead of hours. In parallel, a WhatsApp bot checks in with every instructor days ahead of their session, and the moment someone goes unavailable, it automatically pulls eligible replacements from the roster and hands the PM a ranked, one-tap confirmation list.
// Technical Details
Multi-Modal Session Scoring
Audio and video are pulled from Zoom and Microsoft Teams via a meeting-recording API, transcribed with full speaker-timestamp attribution, and run through an NLP layer that extracts activity signals, comprehension checks, sentiment, and profanity, alongside a vision pipeline sampling video frames for attire, lighting, background, posture, and camera-on status. All nine benchmarks are scored per session and rolled into a color-coded scorecard.
Weekly Reporting
A conversation-heatmap renderer visualizes trainer-student interaction across each session's timeline, and an LLM generates a plain-English narrative summary highlighting each trainer's strongest moment and biggest area to improve — delivered by email or WhatsApp, with timestamped clips for any flagged benchmark.
WhatsApp Availability Management
Every assigned instructor gets automated check-ins three, two, and one day before their session, plus a day-of reminder; silence is treated as unavailability. A detected cancellation triggers an automatic query against the instructor roster for eligible replacements, structured availability requests over WhatsApp, and a ranked confirmation list back to the PM for a one-tap decision.
Interface & Integrations
Google login, automatic roster and session sync to Google Sheets, color-coded trainer health cards (on-track / watch-list / intervention) with one-click drill-down into flagged moments, and a fully containerized deployment.