AWS Partner Network (APN) Blog
Powering a Generative AI Platform Using HAQM DocumentDB with tresle.ai
By Yuxin Yang, Head of AI – tresle.ai
By Sripriya Kannan, Head of Product – tresle.ai
By Ryan Thurston, Head of Go To Market – AWS
By Inderpreet Singh, Sr. Product Manager – AWS
 |
| Tresle |
 |
Increasingly, businesses of all sizes are looking to incorporate generative AI into their operations. Business users with access to large language models (LLMs) benefit from personalized recommendations, near-instant summarization, content generation, and insights into critical operations such as sales forecasting, market trends, or risk assessment. Using generative AI, users can query large, diverse datasets using natural language based on context, not just keywords.
Launched in July 2024, tresle.ai recognizes that few businesses have the technological expertise to build and maintain applications based on generative AI. The company is on a mission to provide a complete generative AI platform that customers can use for virtually effortless integration of generative AI into their applications.
“Businesses have a wave of demand to add generative AI to existing applications but lack the expertise and talent to implement the numerous tools and services required to push a generative AI app into production,” says Patrick Salyer, partner at Mayfield, a Silicon Valley–based venture capital firm that invests in AI. “A need exists to simplify generative AI app implementation by offering a solution that removes all the complexity. tresle.ai provides generative AI integration into existing apps, delivered through APIs, that obfuscates all the complexity of productionizing a generative AI application. This includes data onboarding, choice of a vector database, the embeddings and process of retrieval-augmented generation, model tuning and selection, evaluation, security, and operations.”
Using AWS Managed Services to Build a Plug-and-Play Generative AI Platform
Although tresle.ai launched as a cloud-based startup using more than 49 AWS services, it elected to evaluate all of the market options for its vector database. A critical part of any generative AI solution, vector search helps in capturing the semantic meaning of data by using numerical representations of unstructured data to uncover relationships among disparate datasets, also known as embedding.
“AWS is a state-of-the-art cloud provider, and our team was very familiar with AWS from day one,” says Yuxin Yang, software developer at tresle.ai. “But there are a lot of newer vector databases. We did a rigorous study to choose a vector database, evaluating 10 options across the board.” In making that decision, the company measured more than 20 parameters, such as accuracy, performance, and cost.
After several months of evaluation and testing, tresle.ai elected to build its platform around HAQM DocumentDB, a fully managed native JSON document database that makes it simple and cost-effective to operate critical document workloads at virtually any scale without managing infrastructure. The company benefits from the flexible schema and Mongo DB compatibility of HAQM DocumentDB. The vector search for HAQM DocumentDB feature has the capability to store and query vector embeddings alongside source data, removing the need for separate vector infrastructure and data duplication. Using HAQM DocumentDB, tresle.ai benefits from built-in security practices, data durability, and off-loaded version control and maintenance tasks. “Using HAQM DocumentDB, we don’t have to worry about potential failures,” says Yang. “As an AWS managed service, it saves us a lot of engineering effort.” Additionally, tresle.ai developers appreciated the maturity of the AWS software development kit, which provides tools for tresle.ai to integrate its platform seamlessly with the Rust programming language.
Because HAQM DocumentDB separates compute and storage, tresle.ai’s platform can scale them independently, with the ability to quickly add up to 15 read replicas to handle increased traffic. Moreover, tresle.ai does not need to pre-provision storage, as HAQM DocumentDB automatically scales storage up to 128 TB. “We are able to scale up and down within minutes,” says Yang. “If we were to manage this type of deployment by ourselves, each deployment might take days.”
Furthermore, tresle.ai optimizes the cost of its workload through HAQM DocumentDB I/O-Optimized, a storage configuration for database clusters that provides improved price performance and predictable pricing for customers with input/output (I/O)–intensive applications. Using HAQM DocumentDB I/O-Optimized, Tresle incurs zero charges for read and write I/O operations and thus can more easily predict database spend.
Unlocking a Range of Use Cases through Access to LLMs in HAQM Bedrock
tresle.ai’s platform integrates vector search for HAQM DocumentDB with LLMs accessible in HAQM Bedrock, a fully managed service that provides a single API to access and use various high-performing foundation models from leading AI companies. As a result, tresle.ai customers can search their databases based on meaning, which unlocks a wide range of use cases such as semantic search, product recommendations, personalization, and chatbots. tresle.ai handles the heavy lifting, including connections to the data stores, embedding, and search functionality.
tresle.ai’s customers have the flexibility to onboard any type of data—including PDFs and other unstructured data—and use generative AI capabilities to search and understand context. “We make it easier for our customers,” says Puneet Suri, cofounder of tresle.ai. “Within 1 or 2 days, they have an application that they can play with. We make enterprise applications 50 times faster in implementation at a fraction of the cost.” The customer only needs to select the best model to use within HAQM Bedrock for a particular application.
For example, to generate vector embeddings for unstructured text such as documents, paragraphs, and sentences, tresle.ai uses HAQM Titan Text Embeddings v2, which can intake up to 8,192 tokens and outputs a vector of 1,024 dimensions. Other use cases rely on different models in HAQM Bedrock. For certain applications, such as retrieval-augmented generation applications that are dependent on LLMs, tresle.ai’s platform accesses the latest versions of Anthropic’s Claude in HAQM Bedrock. “Given that we have a small team, we want to focus on our technology edge—our algorithm. We benefit from the accuracy of the information retrieval process and the ability to integrate with a lot of different LLMs,” says Yang. “Using HAQM DocumentDB and HAQM Bedrock together, we are able to scale by minimizing our operational overhead.”
tresle.ai deploys its generative AI platform within the customer’s virtual private cloud or AWS account. Customers benefit from out-of-the-box guardrails and built-in HAQM Bedrock security and privacy controls. They retain full control over their data, which remains secure and private, encrypted in transit and at rest. “Because it’s a managed service, we know the security configurations can’t change that much, which prevents the accidental removal of requirements,” says Yang. “Plus, a lot of our current customers have high compliance considerations, and they want their traffic to stay within their own account.”
Helping More Customers Integrate Generative AI into Their Applications
tresle.ai offers its generative AI platform in AWS Marketplace, where customers can discover, deploy, and manage software that runs on AWS. Meanwhile, the company plans to integrate additional serverless technology as it becomes available, building further efficiencies into the scaling process for enterprise customers.
“Search is a critical part of our platform,” says Suri. “Using HAQM DocumentDB, we offer a comprehensive solution that goes beyond traditional keyword searches. It empowers you with text and semantic search tools, delivering tailored and insightful results.”
Tresle AI – AWS Partner Spotlight
Tresle AI is an AWS ISV Partner delivering an enterprise-grade, production-ready AI Logic Server to seamlessly integrate Gen AI into both existing and new line-of-business applications. Tresle’s AI Logic Server empowers organizations to accelerate innovation, automate complex workflows, and unlock actionable insights. Whether enhancing legacy systems or building next-generation solutions, Tresle makes advanced AI adoption straightforward, reliable, and cost-effective.