← Back
AWS ML Blog

Taxbit Assist: A grounded, role-aware AI agent on Amazon Bedrock AgentCore

•16 min read•
#agents#bedrock#amazon
Level:Advanced
For:AI Engineers
✦TL;DR

Taxbit built “Taxbit Assist,” a role‑aware, citation‑driven AI agent embedded in its dashboard using Amazon Bedrock AgentCore and the Strands Agents SDK. The assistant must ground every answer in a source, enforce organization‑level permissions, avoid providing tax advice, and comply with regional data‑residency and ISO/IEC AI management standards. By leveraging AgentCore’s managed runtime, built‑in memory, and per‑session isolation, the team focused on retrieval pipelines and safety guards, achieving >55 % of users returning within 24 hours for follow‑up queries. This architecture lets compliance‑heavy SaaS products add trustworthy, regulated AI assistants without building their own agent infrastructure.

⚡ Key Takeaways

  • >55 % of users who ask Assist a question return to ask another within 24 hours (internal production analytics).
  • Uses Amazon Bedrock AgentCore runtime with built‑in AgentCore memory for conversation‑state handling.
  • Per‑session isolation enforces role‑based access, preventing support users from seeing admin‑only workflows.
  • Retrieval pipeline supplies cited sources; when no source is found the agent explicitly says it does not know.
  • Compliance built on ISO/IEC 42001, ISO/IEC 27001, ISO/IEC 27701, and SOC 2 Type II certifications.

🔧 Tools & Libraries

Amazon Bedrock AgentCoreStrands Agents SDK
💡 Why It Matters

Engineers can ship regulated, citation‑required AI assistants quickly by offloading agent orchestration, state management, and isolation to Bedrock AgentCore, focusing effort on domain‑specific retrieval and safety—critical for finance and compliance SaaS products.

✅ Practical Steps

  1. Implement the assistant as a Strands Agents SDK application and deploy it on the Amazon Bedrock AgentCore runtime.
  2. Connect the dashboard’s streaming chat API to AgentCore, forwarding user queries for guardrail screening, knowledge‑base retrieval, model reasoning, permission checks, and response streaming.
  3. Continuously measure answer quality on live traffic and feed results back into the retrieval safety pipeline.

Want the full story? Read the original article.

Read on AWS ML Blog ↗

More like this

Introducing workload recovery and expanded infrastructure checks on Amazon SageMaker HyperPod

AWS ML Blog•#deployment

Introducing Gemini 4 Neon

Google DeepMind•#llm

DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

Hugging Face Blog•#llm

Mathematical proof, powered by AI. And no hallucinations!

Amazon Science•#agents

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING