Back to Projects
Agentic AIadvanced

Multi-Agent AI Research Platform with AWS Guardrails, LLM Gateway, Red Teaming, STM, LTM & Semantic Caching

A production-grade autonomous research platform where a 4-agent LangGraph pipeline (Search → Summarize → Write → Verify) processes any topic end-to-end, with every request passing through AWS Bedrock Guardrails, a TensorZero LLM gateway with GPT-4o/Groq fallback, and a three-tier memory system — Redis session memory (STM), pgvector long-term memory (LTM), and semantic caching. Every report is automatically scored by an LLM-as-judge via LangSmith, while a PyRIT red team dashboard continuously stress-tests the system with jailbreak, XPIA, crescendo, and skeleton key attacks to prove the guardrails hold under real adversarial pressure. Full infrastructure on AWS, provisioned with Terraform, deployed via GitHub Actions CI/CD.

21 lectures

What You Will Learn

How to build a production-grade multi-agent AI pipeline — design and wire a LangGraph 4-agent workflow (Search → Summarize → Write → Verify) with an LLM gateway, automatic model fallback, and layered memory (Redis STM, pgvector LTM, semantic cache) that mirrors how real-world AI systems are architected.
How to secure an AI system end-to-end — apply AWS Bedrock Guardrails for input/output safety, implement API authentication and rate limiting, and run automated adversarial red team attacks (jailbreak, XPIA, crescendo, skeleton key) using PyRIT to validate that every guardrail actually holds under real adversarial pressure.
How to ship AI infrastructure like a professional team — provision a complete AWS stack (ECS, RDS, ElastiCache, ALB, Secrets Manager, ECR, VPC) using Terraform, automate build and deployment with GitHub Actions CI/CD including automatic rollback, and monitor every agent run with LangSmith tracing and LLM-as-judge evaluation.

System Architecture

Multi-Agent AI Research Platform with AWS Guardrails, LLM Gateway, Red Teaming, STM, LTM & Semantic Caching Architecture Diagram

High-level architecture overview of the Multi-Agent AI Research Platform with AWS Guardrails, LLM Gateway, Red Teaming, STM, LTM & Semantic Caching .

What You'll Build

  • A 4-agent autonomous research pipeline — a LangGraph-powered system where four specialized agents (Search, Summarize, Write, Verify) collaborate to take any research topic and produce a fully written, verified report — exportable as text, PDF, or structured JSON.
  • A layered AI memory and caching system — a three-tier memory architecture using Redis for short-term session memory (STM), PostgreSQL with pgvector for long-term semantic memory (LTM), and a semantic cache that skips the full pipeline when a similar topic has already been researched — saving cost and latency
  • A live red team security dashboard — a PyRIT-powered attack dashboard that runs four adversarial attack types (jailbreak, XPIA, crescendo, skeleton key) against your own running system, with results showing which attacks were blocked or passed, automated weekly scheduling via EventBridge, and AWS Bedrock Guardrails defending every request in real time.

Project Instructor

Sudhanshu

Sudhanshu

4+ years exp
LinkedIn
Multi-Agent AI Research Platform with AWS Guardrails, LLM Gateway, Red Teaming, STM, LTM & Semantic Caching
Premium
One Subscription. 40+ Projects. Unlimited Access.
AccessMobile & Web