Production-Ready AI Agents: From LLMs to Small Language Models
summary
Roadmap claims a practical migration from frontier LLMs to small language models with concrete results: a hybrid Granite SLM achieves 50% footprint reduction in one deployment, another case cuts cost by 90% and speeds responses 10x, and evaluation runs include 200k automated plus 35k manual checks. Tools and patterns shown include function‑calling, routing and tool‑rack architectures, LangFuse tracing for observability, CI/CD treating prompts as config, test‑driven agent suites, and the Malaya instruct‑validate‑repair approach.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 4:13Production agent requirements
- 9:08Agent architectures and testing
- 18:55Observability and cost
- 24:10CICD and deployment for AI agents
- 26:55Small language models
- 34:19Migrating with SLMs
- 36:04Q&A
related talks
41mThe Fantastic Small Language Models (SLMs) and When to Use Them
40mSurviving the Agentic Hype with Small Language Models
26mLanguage Model Agents Enable Semi-Formal Programming
01h 43mBuilding LLM Agents (Hackathon)
51mTowards a Pragmatic Understanding of Large Language Models
01h 42mBuilding Responsible AI Agents in Decentralized Open-Source Environments
40mAgents in Action: Advancing Open Source AI Missions
46mArchitecting Efficient Agents with Semantic Routing and Mixture-of-Tools