Lex AIMeridianby Lex AI

Build a Multi-Agent Research Assistant

This course has one rule: you build one thing, lesson by lesson. Each lesson adds exactly one capability to the same codebase, and by the end you have a working multi-agent research assistant — a planner that decomposes a question, parallel executors that gather information, shared memory they cooperate through, and a critic that validates the result — built from scratch in Python, running entirely in your browser.

Prerequisites

  • Comfortable Python — functions, classes, list/dict comprehensions
  • A passing acquaintance with async/await (lesson 5 introduces it from zero)
  • No prior LLM or API key required — a deterministic MockLLM runs throughout; you swap in a real model at the end
  • It does not teach LangChain or CrewAI; it teaches the patterns underneath them
Course mapHover any lesson to see why it matters

Prerequisites

Comfortable Python

Functions, classes, list/dict comprehensions

A passing acquaintance with async/await

Lesson 5 introduces it from zero, but it helps to have seen it

No LLM or API key required

A deterministic MockLLM runs throughout; swap in a real model at the end

Lessons

01Beginner

The bare LLM loop

One model, one prompt, one reply — the smallest thing that runs

02Beginner

Give it a tool

The model asks; your code acts and reports back

03Intermediate

ReAct: reason before you act

Interleave a thought and an action on every step

04Intermediate

The Planner: decompose before you search

Break a research question into sub-questions first

05Intermediate

Parallel executors: run sub-questions concurrently

async/await so N sub-questions run at once, not in series

06Intermediate

Shared memory: one source of truth

A store every agent reads from and writes to

07Advanced

The Critic: validate before you deliver

A second agent whose only job is to find the flaws

08Advanced

Close the loop: the Orchestrator

Wire planner, executors, memory, and critic into one system

09Intermediate

Observability: see inside the machine

Trace every step so you can tell which agent is the bottleneck

10Advanced

The whole system

Step back and read the finished assistant as one design

Unlocks

Any agentic framework

LangChain, CrewAI, and the rest read as familiar once you know the patterns underneath

A real-LLM research assistant

Swap the MockLLM for a live model and run the same system for real

Finished this lesson?

Read through the lesson first (0/20s · scrolled 0% of 70%).