How we do AI at Amplitude
summary
Amplitude builds a custom columnar datastore called the Behavior Graph to answer exploratory queries in near real time, runs in-house small models plus fine-tuned LLMs (no massive training bill), and ships features like Data Assist (released July) that scores data governance and surfaces real‑time anomalies so teams act. Prompt engineering is treated as code with an internal experimentation system for prompt variants, and NLP queries via Ask Amplitude map to charts; OpenAI calls send only metadata and require customer opt‑in.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:07Amplitude overview
- 2:52Core analytics & AI
- 6:00Behavior Graph approach
- 8:31AI features & models
- 10:59AI as tool not strategy
- 12:44AI ethics & privacy
- 13:48AI design process
- 14:50Data Assist & governance
- 17:17Prompt engineering challenges
- 19:02Ask Amplitude NLP
- 21:56Future of behavior models
- 22:19Q&A
related talks
30mAI-Driven Operational Data Intelligence
16mGame Data for All: How AI Is Reinventing Game Analytics
28mUnlocking the power of GenAI
27mUsing GenAI with Audio Data for Driving Better Customer Experience
16mForm factors for your new AI coworkers
31mLamRAG: From Data to Constructive Insights using Amazon Bedrock
29mCustomizing genAI applications for your business
17mAI in Production: Lessons Learned