AWS Partner Network (APN) Blog
Driving Innovation in Social Services with Thoughtworks’ GenAI Digital Product Accelerator powered by Amazon Bedrock
By Sarah Sulistio, Product and Innovation Lead – Thoughtworks
By Shipra Shandilya, Community Lead Gurgaon – Thoughtworks
By Philip Tran, Partner Solution Architect – AWS
 |
| Thoughtworks |
 |
Singapore is undergoing a significant social shift, driven by its rapidly aging population. In 2022, 18% of its citizens were seniors aged 65 and above. By 2030, this demographic is projected to increase to 25%. Even with heightened focus and investments in healthcare, economic policies, and social schemes, the social sector grapples with surging healthcare demands and manpower constraints.[1]
Volunteerism is being recognized as an important means to augment these efforts. However, despite its potential, Singapore’s social sector faces significant hurdles. Over half of Singaporeans have expressed an interest to volunteer, but only a mere 22% actually do [2]. A key challenge is the reliance on inefficient processes and inadequate support systems that struggle to attract, retain and seamlessly integrate volunteers. In this context, technological innovation becomes not just beneficial but necessary.
At Thoughtworks, we recognize the immense value generative AI can bring to solving long standing challenges faced by the social sector. This motivated us to explore how AI can reinvent the volunteer experience and amplify its impact towards the aging population.
In partnership with AWS, we designed and developed a Gen AI project by bringing together our expertise in product innovation and AI capabilities and AWS’ advanced infrastructure and solutions for Generative AI. In six weeks, we developed ISLA – an AI-powered chatbot that empowers volunteers to deliver exceptional services to those in need and enables social service providers to streamline their volunteer operations.
In this post, we will explore how we leveraged Amazon Bedrock and Thoughtworks’ GenAI Digital Product Accelerator to define impactful use cases in the volunteer experience, reduce risk in product decisions by assessing desirability, viability, feasibility, and manage ethical and practical considerations of generative AI deployment.
High-Impact AI Use Cases to Streamline the Volunteer Journey
Applying Thoughtworks’ AI-powered Digital Products approach, we began our exploration by identifying high-impact use cases for AI to reduce friction in the volunteer’s journey. We mapped several interactions between volunteers, social service agencies, social workers, and beneficiaries, uncovering significant pain points and opportunities for AI intervention. For each identified area, we envisioned the various roles AI can play: as an Assistant to guide, an Automator to streamline tasks, a Personalizer to tailor experiences, and a Comforter for emotional support.
This strategic alignment of AI capabilities for specific volunteer pain points enabled us to define three clear and impactful use cases for ISLA:
1. Train and onboard new volunteers
Training and onboarding can be a tedious and time-consuming process for both volunteers and social service providers. ISLA summarizes and contextualizes dense and lengthy training materials, responds to queries from volunteers, and provides clear and concise information as they need.
Figure 1 – ISLA can help elevate the onboarding experience for new volunteers.
2. Role-playing to build rapport
For volunteers, building rapport and creating trust with their beneficiaries is crucial to building a healthy partnership. ISLA offers role play simulations, taking on the role of a beneficiaries. This offers volunteers an immersive learning experience that enhances their skills and confidence to build rapport and nurture long-lasting relationships.
3. Simplify documentation through voice-to-text transcription and summarization
Documenting visits and interactions with beneficiaries often takes up a lot of a volunteer’s time, often involving hours of the volunteer’s time to scribe, clarify and condense, and submit their notes. ISLA removes the hassle by simplifying the note-taking process through voice-to-text transcription and summarization. Automating such routine tasks can help free up volunteers so they can focus on what matters most: taking care of their beneficiaries.
Figure 2 – ISLA can quickly summarize documentation using speech-to-text transcription.
From Idea to Impact: Developing ISLA using Amazon Bedrock
Together with AWS, we embarked on a journey to create a proof-of-concept that was technologically advanced but also empathetic and contextually aware. We leveraged Amazon Bedrock to make it simple to evaluate and experiment with top foundation models from leading AI companies like Anthropic, Meta, Stability AI and Amazon, and drew upon Thoughtworks’ expertise in generative AI and deep understanding of the public sector and social services to build a product that showcases our combined expertise and shared commitment to driving innovation and social impact.
An In-Depth Analysis of ISLA’s Technical Architecture
Amazon Bedrock’s Central Role:
- We integrated Amazon Bedrock’s managed services to tap into a diverse array of foundation models, leveraging the service’s capacity to serve large language models (LLMs) through a seamless API interface. This integration is key to scaling our chatbot’s capabilities and making sure that it remains at the forefront of AI-driven innovation.
Strategic Model Selection:
- Our selection of Amazon Titan for the foundational model is a strategic decision that anchors our project in versatility and high performance. Titan’s expansive range of applications spans from nuanced text generation to intricate semantic search capabilities and innovative image generation, all while adhering to the rigorous standards for responsible AI practices shared by both AWS and Thoughtworks.
- The Titan text models are adept at enhancing productivity, enabling a spectrum of tasks including sophisticated content creation, accurate classification, and dynamic conversational exchanges; all pivotal in providing precise, contextually relevant search results that resonate with users.
- Based on our need in terms of allowance, performance and capabilities, Titan offered the most cost-effective solution for the specific use case we had.
- Titan’s model optimization for English, with multilingual support for 100 additional languages (in preview) will allow us to customize the models to our specific organizational needs, striking an optimal balance between performance excellence and cost-effectiveness. This customization is vital for handling extensive datasets and fostering significant user interactions.
Figure 3 – A technical architecture diagram of ISLA.
High-level Workflow:
- Content files and FAQ documents are uploaded, chunked and converted into vector embeddings using Amazon Titan Text Embeddings model. The vector embeddings are stored in a vector DB using Amazon RDS with pgvector extension.
- User types or speaks a question on the ISLA Web interface. If the question is spoken, the Web Speech API converts the audio input into text using Amazon Transcribe.
- The ISLA API performs validation and checks against an in-memory FAQ cache store. If a match is found, the cached answer is returned to the user.
- If there is no match, the question is processed by LangChain’s appropriate question-answering chain. The Retrieval QA performs similarity searches to retrieve the most relevant information from the knowledge base. The search results together with the conversation summary are provided as additional context in the result prompt that is sent to Amazon Bedrock to get the final answer for the user’s question.
Efficient Software Development Kit (SDK) Integration:
- The adoption of the Amazon Bedrock SDK proved to be less challenging than anticipated. Its familiar patterns meant our developers could swiftly adapt to it, leading to rapid development cycles and minimal need for constant documentation consultation.
RAG and Knowledge Base Capabilities:
- The late 2023 release of Amazon Bedrock’s fully-managed RAG service, Knowledge Bases for Amazon Bedrock was promptly explored. At the time of our project implementation, Knowledge Bases supported limited model selection and response customization. Recognizing its potential to streamline our chatbot’s data retrieval process, we decided to keep it on our radar for future integration.
- In the interim, our approach was to leverage LangChain, which already offered robust integration with Amazon Bedrock, to achieve RAG functionality while providing the flexibility we needed for our project requirements.
Crafting Custom Solutions:
- We navigated through the complexities of inter-library dependencies, making sure that our system remains robust against library updates. This approach minimized potential conflicts and preserved the integrity of our chatbot’s operations.
Our exploration and partnership with the AWS team on discussing and debating the landscape of generative AI has been nothing short of exhilarating. The preliminary phases have shone a spotlight on Claude v2.1, favoring its deployment for its remarkable stability and performance – qualities essential for production. Yet, the journey doesn’t end here. We’re energized by the progress of models like Titan and are committed to staying on the cusp of AI innovation. Amazon Bedrock’s seamless integration and rich palette of AI models together with Guardrails for Amazon Bedrock for safeguarding Gen AI applications position it as the ideal fully managed service for our project, making sure that when we shift into production, we’ll have the most sophisticated, managed services at our disposal.
The Future of AI in Social Services
Looking ahead, our vision for ISLA extends far beyond its current capabilities. We envisage it evolving into an AI companion that not only assists but anticipates the needs of volunteers and social workers. The potential to revolutionize the social sector is immense – transforming the way support is delivered, enhancing community engagement, and ultimately, contributing to a more empathetic and effective social services ecosystem. ISLA is just the beginning of a journey towards a smarter, more compassionate future in social care.
Partner with Thoughtworks for Generative AI and Innovation
Embarking on your AI transformation journey with Thoughtworks and Amazon Bedrock is a stride towards pioneering change. We invite enterprises to leverage Thoughtworks’ GenAI Digital Product Accelerator for creating cutting-edge AI solutions. Connect with our team of experts to explore Amazon Bedrock and discover how it can propel your business forward. Further, explore our offerings on the AWS Marketplace to access a suite of tools and services designed for AI innovation. Together, let’s shape a future where technology drives impactful, sustainable growth.
.
.
Thoughtworks – AWS Partner Spotlight
Thoughtworks is an AWS Premier Services Partner and AWS Marketplace Seller with multiple AWS Competencies including Machine Learning, Data & Analytics Services and Government Services. Thoughtworks brings advanced modernization and software engineering expertise to help customers redesign organizations, evolve architecture, and transform business applications and engineering teams.