Build AI Agents

Recently I am thinking about creating workflow supercharged by AI subagents, those agents are work together to get info, distill the key insights, synthesize all infomation. I also want o create a group of agents to form the info about educations, find more materials.

Building a Deep Research

Agent High Level Overview:

1- Orchestrator Agent — Plans research, decomposes query into parallel sub-tasks 2- Perspective Mining — Generate diverse viewpoints before searching (STORM pattern) 3- Parallel Research Agents — Fan-out multiple Tavily searches simultaneously 4- Reflection — Distill each search into reflections, not raw text (~70% token savings) 5- QA/Critic Agent — Verify claims against sources before synthesis 6- Synthesizer Agent — Compile verified findings into final report

Key insight: Parallelize everything (microservices approach). Quality comes from reflection loops, not more prompts.