The Government e-Marketplace (GeM), Indian Government’s online platform for public procurement in the country, has witnessed enormous success, with over 2 million products and the availability of 300 services. GeM has become the second-largest public procurement platform globally, with a gross merchandise value (GMV) of more than $120 billion since its inception. The platform has delivered cost savings of approximately 9-10% on every transaction, resulting in a cumulative savings of around $12 billion for the government treasury. A significant portion of the platform’s success is attributed to its inclusivity, especially for small and medium-sized enterprises (SMEs), which have received $30 billion worth of orders. Recently, at the AWS Empower India Event held at the Bharat Mandapam, Dr Pankaj Dikshit, Chief Technology Officer at the Government e Marketplace (GeM) in his keynote address highlighted how AWS is modernising the platform and its latest offering Gem AI.
Gem AI, powered by Amazon Bedrock, aims to redefine the customer experience on the GeM platform by providing more precise and conclusive answers to queries, unlike the fixed responses previously offered by the GemMy chatbot. “We integrate with over 100 entities and hundreds of APIs, and it’s our mission to ensure that all these integrations are consistently running and that our systems remain available at all times. Given the scale of our operations and our focus on the customer, we face a challenging path of growth. To scale effectively, the best partner for us is, undoubtedly, the cloud. The cloud allows us to quickly set up virtual machines, expand memory on the fly, and meet the demands of scalability. With our customers as a central focus, our vision is clear: we need to digitize, automate, and transform. Additionally, with the emergence of generative AI, we must continue to learn and evolve, repeating this cycle of innovation and improvement,” said Dikshit on the use of Amazon Bedrock to empower GeM AI.
He further went on to add: “Only then can we truly meet the needs and expectations of our users and customers. We streamline processes, improve efficiency, and aim to enhance the user experience on a daily and ongoing basis. Collaboration is key. We actively work with our partners, collaborate internally, and engage with our users to improve our platform. With our users’ best interests in mind, we’ve partnered with AWS over the past few months to develop a generative AI-based chatbot, enhancing the experience for everyone who visits our portal.”
The earlier version of the chatbot was menu-driven, offering a limited range of responses and interactions. “Users were constrained by the fixed options available, making the experience somewhat restrictive. While functional, it lacked flexibility in handling more natural or conversational queries. Users often found themselves limited when trying to express their needs or if they weren’t sure of the exact terms to use,” said Dikshit.
In comparison, the introduction of the new Gem AI chatbot, Dikshit claims, has significantly transformed user engagement. “In contrast, Gem AI offers a much more advanced, user-friendly experience. It supports translation, allows users to input simple English, and even accepts misspelled keywords, all while providing accurate responses. This update reflects a major step forward in enhancing user interaction and accessibility,” explained Dikshit.
Dikshit also stated that the GeM platform has seen consistent traffic, with token usage ranging between 200,000 and 300,000 per day, while user queries reach approximately 1,000 daily. “With the introduction of Gem AI, the question arose: could any large language model (LLM) have been used to build the chatbot? The answer was clear—no. The need was to go beyond a generic conversational engine. The goal was to develop a chatbot tailored to the specific needs of users, one that could provide responses strictly from the GeM (Government e-Marketplace) perspective,” he said.
He further added that to achieve this capability a specialized architecture was required that ensured reliability, security, and the ability to deliver relevant, accurate information. “The chatbot was built with two key components: the LLM for language comprehension, and a carefully curated knowledge base, focused solely on GeM-related content. This setup ensures that users receive responses specific to the government and marketplace, while irrelevant or unrelated queries are automatically filtered out,” reiterated Dikshit.
In addition to providing accurate information, Dikshit said in his address that the team also took measures to minimize hallucinations, which are errors that are common in generative AI models, by validating user inputs and outputs. “For instance, political questions or queries unrelated to the GeM context are rejected, with Gem AI providing a standard response. Looking ahead, the roadmap for JamAI includes significant enhancements. The chatbot is set to offer voice interactions in both English and Hindi, with plans for multilingual support through partnerships with public and private sector solutions. Users will also soon be able to upload screenshots, enabling the system to generate tickets and issue tracking numbers for any problems encountered. This would represent a significant step toward fully automating customer support for the government and marketplace,” said Dikshit.
Other Use Cases GeM is Working on for Gem AI
Diskhit highlighted that they were working on other proof of concepts as far as Gem AI was concerned. “We have more problem cases that we’re working around deliverance. For instance, every morning I receive an email report on our CEO’s desk detailing the previous 24 hours of transactions, volumes, prices, and more. But why does it only come once a day? The answer lies in the sheer scale of the data—our database is nearly a petabyte in size, filled with millions of records. It takes the entire night for scripts and processes to run and generate a report that’s usable by morning.”
Diskhit pointed out the limitations of current processes, explaining how Gem AI could revolutionize this by allowing them to access and generate reports in real-time, with simple English queries. “Instead of waiting for the morning, I should be able to ask for sales figures from the last six hours or a list of the top-selling products at any time of day, and instantly get a response. We should be able to provide a solution where the same report can be requested in simple English, in a conversational manner, understood by the system, and generated instantly.” He explained that they have worked with Amazon’s 6th band and 2nd band as part of their proof of concept (POC) and will continue to do so in the future.
From an AI perspective, Diskhit highlighted the advantages AI offers, such as its ability to read, understand, and process vast amounts of data. He explained how AI could generate various forms of analytics, including descriptive insights and known use cases like collision detection and fraud prevention, leveraging its superior pattern recognition capabilities to offer valuable forecasting and visibility thus providing enriched information for leadership’s strategic planning.
