ML Platform Challenges & Applications at Freshworks
summary
Freshworks runs ML for 200,000 businesses, ingesting ~30 GB/day and keeping 250 million live predictions refreshed in near real time via Kafka consumers that trigger models hosted on SageMaker multimodel endpoints. Architecture uses Spark for text preprocessing, LSTM sequence models for engagement, and classical regressions for static fit; cold-start falls back to industry/global models. Lead scoring is real-time, explains top reasons using LIME/SHAP plus business-rule gating, and surfaces channel/time recommendations.
AI-written summary · about
chapters (5)
- 0:00Introduction to Freshworks
- 2:26Customer engagement AI vision
- 8:50ML platform challenges
- 15:31Predictive lead scoring
- 21:25Next steps and demo
related talks
21mBuilding Scalable ML Systems: Strategies and Insights
27mML Platform Journey
29mBuilding ML Platform Capabilities at Global Scale
30mML Platform for Insurance Conglomerate
24mUnder Pressure: Applying ML in Real Time
26mManaging ML Models @ Scale
32mLiveMLP: ML Platform for Assisting Contact Center Agents in Real-Time
31mUse-Case Driven ML Platforms