self-serve AI analyst
summary
Razorpay’s in-house assistant, DataGaaru, combines a Reasoning Bot that generates and runs SQL against Trino with a Health Metrics Document Generator that runs parallel metric deep-dives; it routes requests across multiple LLMs (including a fast rephraser model) with vector-retrieved golden-query examples, RBAC plus input guardrails for PII, and logging for feedback. Claims: ~1,500+ queries/month, 100+ active users, 100+ concurrent Trino capacity, and ~20% of analytics bandwidth reclaimed.
AI-written summary · about
chapters (11)
- 0:00Razorpay analytics challenges
- 1:29Monday panic slow insights
- 2:52Introducing Data Guru
- 4:17Three innovations for reliability
- 6:27Processing user queries
- 8:12Data security and performance
- 10:25Proving Data Guru works
- 11:51Live demo reasoning bot
- 16:18Document generator health metrics
- 19:17Reasoning engines vs chatbots
- 20:40Q&A segmentation cost
related talks
30mRyoma: Building Open Source AI Agents for Enterprise Data Analysis
30mGenerative AI in action: from POC to business value
45mForget SQL! Create your own AI Data Analyst
18m1,200 analytics requests a day: Clay and Vercel's Bet on AI-Native Analytics
16mTransforming Analytics With AI: A Comprehensive Guide to an Advanced Tech Stack
30mAI-Driven Operational Data Intelligence
14mThe Retail Revolution: Real-World Generative AI Applications
17mUnlocking the Power of AI for Seamless Self-Service BI Transfer