Deploying AI Literature Agents at Scale
summary
Deploys a production RAG stack with domain-adapted LLMs, synchronous embedding pipelines, hierarchical indexing and read replicas to serve 30+ million citations. Engineered for subsecond query latency, GPU-accelerated embedding batches, optimized query caching and staged index updates, it reports ~90% entity recognition accuracy and >85% citation relevance while running A/B experiments and statistical validation. Monitoring surfaces model drift, citation accuracy and resource triggers, and autoscaling plus scheduled scaling yields an 80% reduction in literature search time for target cohorts.
AI-written summary · about
chapters (10)
- 0:00Introduction and Problem Statement
- 1:35AI-Powered Solution Overview
- 2:46ML Ops Architecture
- 4:24Biomedical NLP Pipeline
- 5:59Model Training and Deployment
- 6:46Scaling Vector Databases
- 7:30RAG Architecture and Performance Metrics
- 8:29Monitoring and Observability
- 9:51Continuous Experimentation and Infrastructure Scaling
- 11:17Key Learnings and Conclusion
related talks
13mBuilding High-Performance AI Literature Processing
33mDeep research at scale
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases
46mFullstack AI: The Building Blocks for Production Ready AI-Powered Software
47mReal-Time AI with Open Source
51mTransforming Banking with Advanced AI: GenAI, RAG, and Multi-Agent Systems
28mScaling AI Infrastructure
24mScalable GenAI in Production