Most AI agents today are built around LLMs, but in production that often means higher latency, rising inference costs, and less predictable behavior.

This talk shows a more practical way to build agentic systems using small language models (SLMs). Instead of relying on a single large model for every task, we break the system into modular components powered by smaller, specialized models that handle routing, tool use, structured decisions, and narrow reasoning more efficiently.

Drawing from real-world experience building production AI systems, this session explores how to design agents that are faster, cheaper, and easier to control. We will cover architecture patterns such as specialist model routing, SLM-based tool calling, hybrid SLM-LLM pipelines, and the engineering tradeoffs between cost, latency, accuracy, and operational complexity.

We will also walk through how to deploy SLMs on your own server infrastructure and use Lamatic to build a production-grade agentic system end to end.

If you are building AI-powered products and want to move beyond demo-grade agents, this talk will help you rethink your architecture and design systems that scale in production.

Key takeaways
- How to design agentic systems using small language models
- When SLMs outperform LLMs in real production workflows
- Architecture patterns for faster, cheaper, and more controllable AI agents
- The tradeoffs between latency, cost, capability, and reliability
- How to combine SLMs and LLMs effectively in hybrid systems
- How to deploy SLMs on your own servers and build a production-grade agentic system with Lamatic

Contact Info:
- Portfolio Profile: https://arunaddagatla.vercel.app/
- Phone: +918485019026
- LinkedIn: https://www.linkedin.com/in/arun-addagatla/