LLM agents · CrewAI · OpenAI

Multi-Agent Deep Research System

An AI multi-agent system for automated web research, data retrieval and report generation. Built agent-collaboration workflows using LLM APIs and a tool-calling architecture; deployed on Windows and Linux.

CrewAIOpenAILangChainPythonMulti-Agent

Role: AI engineer
Timeline: 2024

View on GitHub ↗

Field notes hero — a research assistant that argues with itself until it's right, with agent pipeline diagram — Field notes — agent pipeline

Writer agent draft output on on-device LLM inference with numbered findings — Writer agent — draft output

Problem

Manual deep research across the web is slow and hard to reproduce. A single LLM call hallucinates and cannot plan across many sources.

Approach

01Designed a CrewAI multi-agent workflow: planner, retriever, analyst, and writer agents.
02Wired tool-calling for web search, page fetch, and citation collection.
03Persisted intermediate artefacts so runs are auditable and re-runnable.
04Deployed on Windows and Linux with a small CLI + report exporter.

Stack

CrewAI
OpenAI
LangChain
Python
Playwright/Requests

Outcomes

→Reduced research report drafting from hours to minutes.
→Generated structured, cited reports instead of unverified prose.

View on GitHub ↗