AI21 and AWS's Generative AI Hackathon

A hackathon jointly hosted by AI21 and AWS brought together over 50 AWS solutions architects and select GenAI practitioners to tackle real-world challenges such as customer support, grounded content generation, and LLM guardrails. The event was held under the sponsorship of AWS executives.

AI21 explained how generative AI (GenAI) is being used in retail, financial services, healthcare, and more, and shared prompt engineering guides along with methods for mitigating hallucination and inaccuracy. In particular, they emphasized that TSM (Task Specific Models) designed for specific tasks, rather than general-purpose LLMs, are more efficient and reliable, delivering better time-to-value with accurate, data-grounded responses.

The hands-on challenges were as follows.

• LLM Guardrails: Blocking prompt injection attacks and preventing topic drift using blue team and red team approaches
• Grounded Content Generation: Generating commercial content from source data such as product data
• Enhanced Customer Support: Using past resolved tickets to assist with responses to current cases
• Multi-Document Summarization: Structured summarization across scattered documents and large corpora
• Ad Creation: Automating ideation, content generation, localization, targeting, and social integration
• Call Transcript Summarization: Consistent summarization and information extraction from conversation records

The key was RAG (Retrieval Augmented Generation), which grounds responses in enterprise data. The Contextual Answers TSM, used together with the Jurassic-2 LLM in customer chatbots, can search an organization's knowledge base and then provide data-grounded responses, and AI21's models were also integrated with AWS-native services such as Amazon Bedrock, Amazon SageMaker, Amazon OpenSearch, Amazon Lex, and AWS Lambda.