Chainly Asia: Achieving 35% Infrastructure Cost Reduction at Scale
PROJECTS
GNXT
7/28/20265 min read
Customer Challanges
Technologies
Furthermore, traditional keyword search mechanisms were incapable of comprehending complex, domain-specific semantics across heterogeneous trade documents written in multiple languages. When operational disruptions or customs holds occurred, retrieving relevant clauses or historical resolution records required days of manual document searching, directly threatening customer Service Level Agreements (SLAs) and inflating operational overhead. To maintain its market leadership and unlock scalable growth, Chainly Asia urgently required an enterprise-grade, highly secure Generative AI platform capable of automating semantic document analysis, enabling instant natural language decision support, and accelerating multi-tenant trade compliance verification—all while strictly maintaining data privacy and regional security regulations.
About The Customer
Chainly Asia is a premier supply chain technology and logistics visibility platform enterprise based in the Asia-Pacific region. The organization delivers digital infrastructure that unifies multi-enterprise logistics data, enables end-to-end shipment traceability, and streamlines international trade documentation for global manufacturers, freight forwarders, and regulatory bodies. Operating across complex cross-border trade corridors, Chainly Asia processes high volumes of operational transactions and critical supply chain paperwork daily, empowering enterprise clients to convert fragmented operational data into actionable intelligence and transparent supply chain workflows.
As Chainly Asia expanded its international logistics network, the volume and operational complexity of unstructured supply chain documentation—such as international bills of lading, customs clearance filings, multi-currency commercial contracts, and compliance audit reports—increased exponentially. The company relied heavily on manual data verification and legacy relational database queries to process operational inquiries and execute regulatory compliance checks. Freight operations teams and compliance officers were required to spend weeks manually cross-referencing trade agreements, verifying commodity HS codes, and compiling compliance risk assessment documentation. This manual processing created severe operational bottlenecks, extended shipment clearance cycles, and increased exposure to human error and compliance penalties.
GNXT Solutions
Foundation Model Selection, Advanced RAG, and Responsible AI Implementation
Genextech Solutions (GNXT) initiated the engagement by implementing the AWS Generative AI Adoption Framework to systematically evaluate Chainly Asia’s data maturity, security posture, and organizational readiness. Recognizing the critical sensitivity of cross-border trade records, GNXT conducted a comprehensive data governance assessment and risk analysis to establish robust boundary controls. Following a phased implementation roadmap, GNXT validated the core technology stack through a rapid Proof of Concept (POC) focusing on semantic search and automated document summarization.
Adoption Strategy, Readiness Evaluation, and Architectural Foundation
Aligning strictly with the AWS Well-Architected Framework, GNXT engineered a robust security and operational architecture. Network boundaries are isolated within a private VPC, with all network traffic between AWS Fargate, Lambda, and Amazon Bedrock routed securely through AWS PrivateLink and VPC Endpoints, bypassing the public internet entirely. Incoming web traffic is safeguarded by AWS WAF policies attached to Amazon API Gateway. All customer data at rest and in transit is encrypted using customer-managed keys in AWS KMS, while dynamic data anonymization procedures redact sensitive commercial details before model inference.
Upon successful POC validation, GNXT architected a multi-tenant, serverless production environment engineered for continuous scalability, high availability, and operational resilience. The web user interface and administrative management portals were deployed on AWS Fargate, a fully managed serverless container service that scales computing resources dynamically based on incoming user traffic, eliminating the operational overhead of managing underlying physical or virtual server infrastructure.
Production Operations, Enterprise Security, and Lifecycle Support
To deliver high-precision document intelligence, GNXT evaluated multiple Large Language Models (LLMs) based on context window capacity, semantic accuracy, inference latency, and cost-efficiency. GNXT selected Anthropic Claude 3 models hosted on Amazon Bedrock as the core foundation model due to their exceptional reasoning capabilities and support for extensive 200k+ token context windows necessary for complex, multi-page legal and trade documents. For vector indexing and semantic retrieval, GNXT deployed Amazon OpenSearch Service, creating a high-performance vector database to store and query high-dimensional document embeddings generated via Amazon Bedrock Titan Text Embeddings.
GNXT implemented an advanced Retrieval-Augmented Generation (RAG) pipeline utilizing the LlamaIndex framework. Document ingestion is fully automated: when users upload unstructured files, an event-driven workflow triggers Amazon EventBridge to execute AWS Lambda functions that segment documents, extract metadata, generate embeddings, and index document chunks into Amazon OpenSearch. To ensure domain-specific accuracy, GNXT engineered sophisticated Few-Shot and Chain-of-Thought prompt templates. Furthermore, GNXT integrated Amazon Bedrock Guardrails to enforce responsible AI practices. The guardrails continuously filter Personally Identifiable Information (PII), suppress model hallucinations, restrict responses strictly to verified supply chain documentation, and ensure complete compliance with corporate governance policies.
For continuous operational excellence, GNXT established automated Infrastructure as Code (IaC) deployment pipelines using AWS CDK and GitHub Actions, enabling zero-downtime updates and rapid multi-tenant provisioning. System performance, API latencies, and token utilization are continuously monitored through Amazon CloudWatch dashboards with automated alert thresholds. A Human-in-the-Loop (HITL) workflow routes low-confidence model responses to compliance experts for manual verification. Post-launch, GNXT provided a 3-month stabilization period alongside a tiered 24/7 SLA support agreement (guaranteeing sub-1-hour response times for critical operational issues) and quarterly review cycles to iteratively refine model performance based on real-world user feedback.
Architecture Overview & Component Breakdown
End-to-End Technical Workflow Narrative
The production architecture is built as a cloud-native, serverless multi-tenant environment on AWS, adhering strictly to the Security and Operational Excellence pillars of the AWS Well-Architected Framework.
User Request & Ingestion Layer: External user queries and document upload requests pass through AWS WAF firewall policies into Amazon API Gateway, which forwards application traffic to containerized web UI services running on AWS Fargate. Uploaded trade documents are securely stored in private Amazon S3 buckets configured with S3 VPC Endpoints.
Event Processing & Indexing Layer: Document upload events automatically emit signals to Amazon EventBridge. EventBridge rules trigger AWS Lambda execution functions equipped with LlamaIndex parsers. The Lambda functions extract text chunks, send request payloads to Amazon Bedrock Titan Text Embeddings, and write the generated vector embeddings into Amazon OpenSearch Service.
Retrieval-Augmented Generation (RAG) Layer: When a user submits a natural language query, AWS Fargate executes a semantic vector query against Amazon OpenSearch Service to retrieve the top matching document context chunks. Fargate constructs a contextualized prompt combining the user query, retrieved document chunks, and domain-specific system instructions, and transmits it to Anthropic Claude on Amazon Bedrock via private AWS PrivateLink endpoints.
Safety Verification & Delivery Layer: Before the final answer is rendered to the user, Amazon Bedrock Guardrails evaluate the prompt and generated response in real time to filter out PII, verify factual alignment, and block unauthorized topics. The validated output is sent back to the user interface on AWS Fargate.
Cross-Cutting Management Subsystems: AWS KMS manages cryptographic keys for all stored data, AWS IAM enforces role-based least-privilege policies across all component interactions, and Amazon CloudWatch aggregates system performance telemetry and execution logs.
About GNXT
Genextech Solutions is a next-generation technology consulting firm and AWS Advanced Tier Services Partner. We specialize in AI, Cloud Innovation, DevOps, and Automation — helping businesses modernize infrastructure, accelerate digital transformation, and unlock sustainable growth.
We deliver practical, secure, and scalable technology solutions tailored to your business goals. More details
GENEXTECH SOLUTIONS
Empowering businesses with innovative cloud technology solutions
© 2025. All rights reserved.
Address
17B, Street 12C, Long Binh Ward
HCMC, Vietnam
Contacts
+84 38 569 4657
innovate@genextech-solutions.com