BR Academy

Building AI Agents: Tool Use, RAG and Multi-Agent Systems

Exit course

← Building AI Agents: Tool Use, RAG and Multi-Agent Systems

1Foundations u2013 LLM Capabilities and Limits0/3

LLM APIs and token economics15m
Hallucination, reasoning gaps, and when agents help
The planning loop, tool use, and memory
Module quiz

2Tool Use and Function Calling0/4

OpenAI, Anthropic, and open-source function schemas
Forcing JSON responses and validating agent decisions
Wrapping APIs, databases, and code execution safely
Sequential, parallel, and conditional tool chains
Module quiz

3Retrieval-Augmented Generation (RAG)0/5

Semantic, sliding-window, and recursive chunking
Choosing models, dimensionality, and similarity search
Pinecone, Weaviate, Milvus, and local alternatives
BM25, dense retrieval, and cross-encoder reranking
Measuring retrieval quality and end-to-end accuracy
Module quiz

4Memory, Planning, and Multi-Agent Orchestration0/4

Short-term, long-term, and episodic memory patterns
ReAct, chain-of-thought, and tree-of-thought
Hierarchical, peer-to-peer, and market-based orchestration
Routing decisions, context passing, and state management
Module quiz

5Evaluation, Cost Control, and Observability0/3

Unit tests for agents, golden datasets, and metrics
Token budgets, caching, batching, and model selection
Tracing agent calls, debugging failures, and monitoring drift
Module quiz

6Safety, Security, and Deployment0/4

Attack vectors and input validation strategies
Filtering unsafe responses and enforcing constraints
FastAPI, async patterns, rate limiting, and scaling
Monitoring, alerting, cost tracking, and user feedback loops
Module quiz

7Capstone u2013 Build and Ship a Multi-Agent Service0/1

Design and deploy a live multi-agent system
Module quiz

Evaluation, Cost Control, and Observability·Lesson 18 of 24·18 min

Token budgets, caching, batching, and model selection

This lesson is for enrolled learners

Concept Module 1 established that every agent decision is also a cost decision. This lesson turns that principle into concrete techniques for controlling cost and latency once an agent is handling real traffic, not just a handful of test calls. Token budgets set explicit per-request and per-user ceilings — a maximum number of planning-loop steps (from Module 1's step limit),……

Enroll to continue Already enrolled? Sign in