DTDC is India’s main built-in specific logistics supplier, working the most important community of buyer entry factors within the nation. DTDC’s technology-driven logistics options cater to a variety of consumers throughout numerous {industry} verticals, making them a trusted companion in delivering excellence.
DTDC Categorical Restricted receives over 400,000 buyer queries every month, starting from monitoring requests to serviceability checks and transport charges. With such a excessive quantity of shipments, their current logistics agent, DIVA, was operated on a inflexible, guided workflow, forcing customers to observe a structured path fairly than partaking in pure, dynamic conversations. The shortage of flexibility resulted in elevated burden on buyer help groups, longer decision occasions, and poor buyer expertise.
DTDC was searching for a extra versatile, clever assistant—one that might perceive context, handle complicated queries, and enhance effectivity whereas lowering reliance on human brokers. To attain a greater buyer expertise, DTDC determined to boost DIVA with generative AI utilizing Amazon Bedrock.
ShellKode is an AWS Companion, born-in-the-cloud firm specializing in modernization, safety, knowledge, generative AI, and machine studying (ML). With a mission to drive transformative development, ShellKode empowers companies by means of state-of-the-art expertise options that deal with complicated challenges and unlock new alternatives. Utilizing deep {industry} experience, they ship tailor-made methods that foster innovation, effectivity, and long-term success in an evolving digital panorama.
On this publish, we focus on how DTDC and ShellKode used Amazon Bedrock to construct DIVA 2.0, a generative AI-powered logistics agent.
Answer overview
To handle the restrictions of the present logistics agent, ShellKode constructed a complicated agentic assistant utilizing Amazon Bedrock Brokers, Amazon Bedrock Information Bases, and an API integration layer.
When prospects work together with DIVA 2.0, they expertise a seamless, conversational interface that understands and responds to their queries naturally. Whether or not monitoring a bundle, checking transport charges, or inquiring about service availability, customers can ask questions in their very own phrases with out following a inflexible script. DIVA 2.0’s enhanced AI capabilities permit it to know context, handle complicated requests, and supply correct, customized responses, considerably enhancing the general buyer expertise and lowering the necessity for human intervention. The next high-level structure diagram illustrates the appliance stream and the answer structure with AWS companies.
The DTDC logistics agent is designed utilizing a modular and scalable structure to offer seamless integration and excessive efficiency. This streamlined workflow demonstrates how a generative AI-powered serverless logistics agent utilizing AWS App Runner, Amazon Bedrock Brokers, AWS Lambda, and a vector-based information base handles consumer queries starting from monitoring requests to serviceability checks and transport charges intelligently and effectively.
The logistics agent is hosted as a static web site utilizing Amazon CloudFront and Amazon Easy Storage Service (Amazon S3). The logistics agent is built-in with the DTDC web site, which gives an intuitive and user-friendly interface for end-user interactions (see the next screenshot).
An end-user accesses the logistics agent by means of the DTDC web site and submits queries like monitoring shipments, checking service availability, calculating transport charges, FAQs, and so forth utilizing pure language.The consumer requests are processed by App Runner, which helps run the net software (together with API companies, backend net companies, and web sites) on AWS. App Runner is hosted with a number of API companies, such because the Amazon Bedrock Brokers API and Dashboard API. App Runner initiates the Amazon Bedrock Brokers API based mostly on the consumer requests.
Amazon Bedrock is a completely managed service that gives a selection of {industry} main basis fashions (FMs) together with a broad set of capabilities to construct generative AI functions, simplifying improvement with safety, privateness, and accountable AI. With Amazon Bedrock, your content material isn’t used to enhance the bottom fashions and isn’t shared with any mannequin suppliers. Amazon Bedrock Guardrails gives configurable safeguards to assist safely construct generative AI functions at scale. To be taught extra, see Construct protected and accountable generative AI functions with guardrails. AWS Id and Entry Administration (IAM) helps directors securely management who will be authenticated and licensed to make use of Amazon Bedrock sources.
The Amazon Bedrock brokers are configured in Amazon Bedrock. An Amazon Bedrock agent receives the request and interprets the consumer’s intent utilizing its pure language understanding capabilities. Based mostly on the interpreted intent, the agent triggers an acceptable Lambda operate, reminiscent of:
- Monitoring consignments
- Pricing data
- Location serviceability examine
- Assist ticket creation
The triggered Lambda operate calls the next shopper APIs, retrieves the related knowledge, and returns the response to the agent:
- Monitoring System API – Retrieves real-time standing and gives updates on consignment cargo monitoring
- Supply Franchise Location API – Checks the service availability to ship the parcels between the areas
- Pricing System API – Calculates the transport charges based mostly on cargo particulars supplied by the consumer
- Buyer Care API – Creates a help ticket for the end-users
The agent passes the response to the massive language mannequin (LLM), on this case Anthropic’s Claude 3.0 on Amazon Bedrock, which understands the context of the retrieved knowledge, processes it, and generates a significant response for the consumer.
The information base comprises web-scraped content material from the DTDC web site, inside help documentation, FAQs, and operational knowledge, enabling real-time updates and correct responses. The information base contents are saved as vector embeddings in Amazon OpenSearch Service, offering fast and related responses. For normal queries, the logistics agent fetches data from Amazon Bedrock Information Bases, offering accuracy and relevance. Utilizing semantic similarity search, related chunks of knowledge are retrieved from the information base based mostly on the consumer’s question, which Amazon Bedrock then makes use of to generate a context-aware response. If no related knowledge is discovered within the information base, a fallback response (preconfigured within the Amazon Bedrock immediate) is returned, indicating that the system couldn’t help with the request.
The logistics agent queries and related responses are saved in Amazon Relational Database Service (Amazon RDS) for PostgreSQL for enhanced scalability and relational knowledge dealing with. App Runners initiates the Dashboard API name to replace the queries and related responses in Amazon RDS. We focus on this in additional element the next part.
All through the method, Amazon CloudWatch Logs captures key occasions reminiscent of intent recognition, Lambda invocations, API responses, and fallback triggers for auditing and system monitoring. AWS CloudTrail data and displays exercise within the AWS account, together with actions taken by customers, roles, or AWS companies. It logs these occasions, which can be utilized for operational auditing, governance, and compliance.
Amazon GuardDuty is a menace detection service that repeatedly displays, analyzes, and processes AWS knowledge sources and logs in your AWS surroundings. GuardDuty makes use of menace intelligence feeds, reminiscent of lists of malicious IP addresses and domains, file hashes, and ML fashions to determine suspicious and probably malicious exercise within the AWS surroundings.
Logistics agent dashboard
The next high-level structure diagram illustrates the logistics agent dashboard, which captures the end-user interactions and its related responses.
The logistics agent dashboard is hosted as a static web site utilizing CloudFront and Amazon S3. Dashboard entry is allowed just for the DTDC admin crew.
The dashboard is populated by means of API calls utilizing Amazon API Gateway with Lambda as a backend, which retrieves the dashboard knowledge from Amazon RDS for PostgreSQL.
The dashboard gives real-time insights into the logistics agent efficiency, together with accuracy, unresolved queries, question classes, session statistics, and consumer interplay knowledge (see the next screenshot). It gives actionable insights with options reminiscent of warmth maps, pie charts, and session logs. Actual-time knowledge is logged and analyzed on the dashboard, enabling steady enchancment and fast situation decision.
Answer challenges and advantages
When implementing DIVA 2.0, DTDC and ShellKode confronted a number of important challenges. Integrating real-time knowledge from a number of legacy programs was essential for offering correct, up-to-date data on monitoring, charges, and serviceability. This was doubtless addressed by means of the strong API integration capabilities of Amazon Bedrock Brokers. One other main hurdle was coaching the AI to know complicated logistics terminology and multi-step queries, which was overcome through the use of Amazon Bedrock LLMs and Amazon Bedrock Information Bases, fine-tuned with industry-specific knowledge. The crew additionally needed to navigate the fragile means of transitioning from the previous inflexible DIVA system whereas sustaining service continuity and preserving historic knowledge, probably using a phased strategy with parallel programs. Lastly, scaling the answer to deal with over 400,000 month-to-month queries whereas sustaining efficiency was a major problem, addressed through the use of the cloud infrastructure of Amazon Bedrock Brokers for optimum scalability and efficiency. These challenges underscore the complexity of upgrading to an AI-powered system in a high-volume, data-intensive {industry} like logistics, and spotlight how AWS options supplied the mandatory instruments to beat these obstacles. DTDC realized the next advantages from powering the logistics agent with generative AI utilizing Amazon Bedrock:
- Enhanced conversations and real-time knowledge entry with buyer help brokers – Powered by Amazon Bedrock Brokers, the answer improves pure language understanding, enabling extra fluid and fascinating conversations. With multi-step reasoning, it will probably deal with a broader vary of queries with higher accuracy. Moreover, by integrating seamlessly with DTDC’s API layer, the logistics agent gives real-time entry to very important data, reminiscent of monitoring shipments, service availability, and calculating transport charges. The mix of superior conversational capabilities and real-time knowledge gives quick, correct, and contextually related responses.
- Clever knowledge processing and correct FAQ responses – For complicated queries, the logistics agent makes use of LLM expertise to course of uncooked knowledge and ship structured, tailor-made responses. This makes positive customers get clear, actionable insights. For ceaselessly requested questions, the logistics agent makes use of Amazon Bedrock Information Bases to ship exact solutions with out requiring human help, lowering wait occasions and enhancing the general consumer expertise.
- Diminished dwell agent dependency and steady enchancment – Though the logistics agent hasn’t eradicated the necessity for buyer help, the variety of queries dealt with by the client help crew has diminished by 51.4%. The system gives useful insights into key efficiency metrics like peak question occasions, unresolved points, and general engagement by means of built-in real-time analytics, serving to refine and enhance the assistant’s capabilities over time.
Outcomes
The generative AI-powered logistics agent has diminished the burden on buyer help groups and shortened decision occasions, leading to higher buyer expertise:
- Powered by Amazon Bedrock, DIVA 2.0 understands queries in pure language and helps dynamic conversations with a response accuracy of 93%
- Based mostly on the final 3 months of dashboard metrics knowledge, they noticed the next:
- 71% of the inquiries had been associated to consignments (256,048), whereas 29.5% had been normal inquiries (107,132)
- 51.4% of consignment inquiries (131,530) didn’t end in a help ticket, whereas 48.6% (124,518) led to new help ticket creation
- Of the inquiries that resulted in tickets, 40% began with the client help heart earlier than transferring to the AI assistant, whereas 60% started with the assistant earlier than involving the client help heart
DIVA 2.0 has diminished the variety of queries dealt with by the client help crew by 51.4%. DTDC’s help crew can now deal with extra important points, enhancing general effectivity.
Abstract
This publish demonstrated how Amazon Bedrock can remodel a conventional chatbot to a generative AI-powered logistics agent that gives higher buyer expertise by means of dynamic dialog. For companies dealing with comparable challenges, this resolution gives a blueprint for modernizing your AI assistant whereas sustaining compliance with {industry} requirements.
To be taught extra about this AWS resolution, contact AWS for additional help. AWS can present detailed details about implementation, pricing, and the best way to tailor the answer to your particular enterprise wants.
Concerning the authors
Rishi Sareen – Chief Info Officer (CIO), DTDC is a seasoned expertise chief with over twenty years of expertise in driving digital transformation, enterprise IT technique, and innovation throughout the logistics and provide chain sector. He focuses on constructing agile, AI-driven, and safe expertise ecosystems that improve operational effectivity and buyer expertise. Rishi leads initiatives spanning system modernization, knowledge intelligence, automation, cybersecurity, cloud, and synthetic intelligence. He’s deeply dedicated to aligning expertise with enterprise outcomes whereas fostering a tradition of steady enchancment and purposeful innovation. A robust advocate for people-centric management, Rishi locations excessive emphasis on nurturing expertise, constructing high-performing groups, and mentoring future-ready expertise leaders who can thrive in dynamic, AI-powered environments. Identified for his strategic imaginative and prescient and disciplined execution, he has led large-scale digital initiatives and transformation applications that ship lasting enterprise influence.
Arunraja Karthick – Head – IT Providers & Safety (CISO), DTDC is a strategic IT and cybersecurity chief with over 15 years of expertise driving enterprise-scale digital transformation. Because the Head of IT Providers & Safety (CISO) at DTDC Categorical Restricted, he leads the group’s core IT, cloud, and safety applications—reworking legacy environments into agile, safe, and cloud-native ecosystems. Beneath his management, DTDC has adopted a hybrid cloud structure spanning AWS, GCP, and on-prem colocation, with a imaginative and prescient to allow dynamic workload mobility and vendor-neutral scalability. Arunraja has led important modernization efforts, together with the migration of key enterprise functions to microservices and containerized platforms, whereas making certain excessive availability and regulatory compliance. Identified for his deep technical perception and execution self-discipline, he has carried out enterprise-wide cybersecurity frameworks—from E mail DLP, Cell Machine Administration, and Conditional Entry to Hybrid WAF and superior SOC operations. He has additionally championed safe entry transformation by means of Zero Belief-aligned Safe WebVPN, redefining how inside customers entry company apps. Arunraja’s management is grounded in platform considering, automation, and a user-first mindset. His current initiatives embody the enterprise rollout of GenAI copilots for buyer expertise and operations, in addition to unified policy-based DLP and content material management mechanisms throughout endpoints and cloud. Acknowledged as an Influential Expertise Chief, Arunraja continues to problem typical IT boundaries—aligning safety, agility, and innovation to energy enterprise evolution.
Bakrudeen Ok an AWS Ambassador, leads the AI/ML observe at Shellkode, specializing in driving innovation in synthetic intelligence, particularly in Generative AI. He performs a key function in constructing groups and superior AI options, Agentic Assistants, and different next-gen applied sciences. Bakrudeen has made notable contributions to AI/ML analysis and improvement. In 2023 and 2024, he obtained the Generative AI Consulting Excellence Companion Award on the AI Conclave and the Social Affect Companion of the 12 months Award for Generative AI at AWS re:Invent 2024, each on behalf of Shellkode reflecting the crew’s sturdy dedication to innovation and influence within the AI area.
Suresh Kanniappan is a Options Architect at AWS, dealing with Automotive, Manufacturing and Logistics enterprises in India. He’s keen about cloud safety and Trade options that may resolve actual world issues. Previous to AWS, he labored for AWS prospects and companions in consulting, migration and resolution structure roles for over 14 years.
Sid Chandilya is a Sr. Buyer Relations Supervisor at AWS, answerable for tech led enterprise transformation with Automotive, Manufacturing and Logistics enterprises in India. Sid is peculiarly keen about difficult standing quos, constructing a joint “Suppose Huge” imaginative and prescient with buyer CXOs and leveraging Ai infused tech to speed up outcomes. He’s identified for his deep understanding of {industry} imperatives (working backward from buyer) and translating the enterprise ache factors into tech resolution.
