top of page
estrellas 2.png

Empowering Legal Professionals:
The client’s Agentic AI Case Intelligence on AWS

About the client

The client is a SaaS platform for legal practice management that centralizes legal proceedings and integrates with judicial and administrative portals such as the National Judicial Portal (PJN). The agentic implementation extends this platform with conversational access to operational and judicial case information for attorneys and internal legal staff.

Challenges

The client needed to replace manual court-record review and spreadsheet-based reporting with a conversational analytical agent capable of retrieving, understanding, and processing legal case data. Slow information dissemination and manual review delayed access to case status, dates, updates, and jurisdictional information and increased the operational effort required from legal teams. Without intervention, the client faced continued operational cost, limited scalability of legal support and analysis, and a higher risk of inaccurate or outdated case interpretation. We identified this as an Agentic AI use case because the required solution needed to understand analytical requests, access external and internal legal data sources, invoke tools, and execute structured retrieval flows over live business information.

Project Goals & Objectives

  • Primary Goal: Deploy a production conversational legal agent using Amazon Bedrock Agents to answer analytical questions about legal cases.

  • Reduce the operational workload associated with manual court-record review and spreadsheet-based reporting.

  • Improve accuracy and contextual relevance of automated case responses.

  • Expand the information sources available to users by integrating judicial and operational data.

  • Provide secure role-based access and production observability for approximately 100 internal users.

The Solution

We implemented the client in production using Amazon Bedrock Agents with Claude 3.0 Sonnet as the final foundation model. We evaluated Claude 3.0 Haiku during validation but selected Sonnet because it provided stronger accuracy, latency, and integration performance for legal case retrieval and analytical queries. A single Bedrock Agent uses Action Groups and prompt engineering to invoke tools and execute structured flows over PJN and the client operational data. Amazon API Gateway and AWS Lambda provide the integration path to PJN and MongoDB Atlas, while Amazon ECS with AWS Fargate supports the managed application runtime. We deployed the solution using AWS CDK.

Designation Definition Fit

This reference is a strong fit for the AWS Agentic AI specialization because we deployed a production Amazon Bedrock Agent that interprets legal analytical requests, determines and invokes the required Action Group tools, retrieves information from external judicial and internal operational systems, and returns grounded case intelligence. The workflow demonstrates autonomous tool use and structured execution over real business data rather than standalone text generation. It also demonstrates production security, observability, role-based access, and responsible AI controls around an operational agent used by approximately 100 internal users.

Architecture-design

Key Technologies Implemented

🔹Amazon Bedrock Agents - production managed agent orchestration. | 🔹Claude 3.0 Sonnet on Amazon Bedrock - selected after evaluation against Claude 3.0 Haiku. | 🔹Bedrock Agent Action Groups - tool invocation and structured execution flows. | 🔹Amazon API Gateway + AWS Lambda - integration with PJN and MongoDB Atlas. | 🔹Amazon ECS with AWS Fargate - managed production application runtime. | 🔹MongoDB Atlas - operational legal data and low-latency access. | 🔹Amazon S3 + AWS Glue - external legal data ingestion/storage path documented in the architecture. | 🔹AWS KMS - encryption/security control confirmed in production. | 🔹 AWS CloudTrail - audit logging confirmed in production.  |  🔹Amazon CloudWatch - monitoring/observability confirmed in production. | 🔹AWS CDK + AWS SDKs - deployment and application integration tooling.

Technical Solution & Justification

We designed the production solution by combining Bedrock Agents with Action Groups and serverless/API integrations to provide secure tool-based interoperability with PJN and MongoDB Atlas. Native application authentication and role-based authorization restrict access according to legal team and user profile permissions. MongoDB Atlas connectivity uses an IP allowlist. KMS, CloudTrail, and CloudWatch provide security and observability controls. Responsible AI measures include explicit non-legal-advice disclaimers, prompt testing, customer validation, and factual-accuracy review. The production solution uses managed scaling behavior and no caching layer while achieving response times under two seconds for case-status queries.

Architecture Highlights

  • Production agentic workflow: a Bedrock Agent invokes Action Groups over judicial and operational data.

  • External interoperability: PJN and MongoDB Atlas are accessed through API Gateway and Lambda.

  • Security and governance: role-based authorization, KMS, CloudTrail, CloudWatch, and MongoDB IP allowlisting.

  • Responsible AI: non-legal-advice disclaimers, factual accuracy review, prompt testing, and customer validation.

Solution Optimality & Alternative Approaches

We selected Amazon Bedrock Agents to replace a manual analytical workflow with a managed agent capable of tool invocation and structured legal-data retrieval. We selected Claude 3.0 Sonnet over Haiku after validation because it delivered better performance for legal case retrieval, understanding, and analytical question answering. The managed AWS runtime reduced infrastructure overhead while supporting production use for approximately 100 internal users.

Business Impact & Key Performance Indicators (KPIs)

We measured the following KPIs to validate the business impact of the implemented solution:
 

  • KPI 1 - Response Time: Baseline: the absolute pre-deployment average is not stated in the supplied materials. Improvement target: reduce the time required to answer legal inquiries and case-status questions. Actual result: 30% reduction in response time; the production agent also achieved case-status response times under two seconds. Measurement method: compare average response times before and after deployment.

  • KPI 2 -  Response Accuracy: Baseline: the absolute pre-deployment accuracy score is not stated in the supplied materials. Improvement target: improve the accuracy and contextual relevance of automated legal responses. Actual result: 25% increase in response accuracy. Measurement method: user feedback and accuracy tests against known legal scenarios.

Unmet Goals & Lessons Learned

The production rollout reinforced the importance of iterative prompt testing, customer validation, and factual-accuracy review for legal-data workflows. We also identified continued expansion of legal knowledge coverage as an improvement area. The combination of explicit non-legal-advice disclaimers, role-based access, operational monitoring, and customer validation allowed us to improve usefulness while keeping the agent within its intended operational scope.

Get in touch

Take your business to the next level.

bottom of page
window.addEventListener('load', function() {   var search = window.location.search;   if (!search || search === '?') return;   var params = search.slice(1);   setTimeout(function() {     document.querySelectorAll('iframe').forEach(function(fr) {       try { fr.contentWindow.postMessage({type:'TERACLOUD_UTM', params: params}, '*'); } catch(e) {}     });   }, 1500); });