February 2025 Summaries

PromptQL is a data agent that enables building AI assistants with natural language capabilities, connecting directly to existing data sources while maintaining granular role-based authorization and security. It uses AI to create structured query plans, executes them programmatically outside the LLM's context window, stores results in structured artifacts, and combines retrieval, computation, and AI reasoning in a predictable way. This approach delivers 5x better performance than traditional methods, with near-perfect repeatability even as tasks become more complex. PromptQL showcases its capabilities through a customer support assistant demo, demonstrating how it can query databases, access customer support tickets, and take actions using natural language, all set up in under 5 minutes.

PromptQL is a data access agent that integrates web search, allowing users to enrich private data with real-time public information. This enables businesses to supplement customer records with up-to-date details from the web, opening up various use cases. The author explores different web search tools, including traditional engines like Brave Search and DuckDuckGo, as well as AI-powered retrieval systems such as Perplexity and Google Gemini Grounded with Google Search. They provide a comparison of these tools, highlighting their strengths and limitations, and offer quickstart guides for integrating them into PromptQL setups. By leveraging these connectors with custom business logic, users can enrich private data with up-to-date public information, choosing between pure search for direct keyword results or AI-powered retrieval for synthesized insights.

Hasura's metadata-based approach provides an elegant solution for financial institutions to address data governance challenges, ensuring security, compliance, and transparency while making their enterprise more AI-ready. By automatically generating APIs from data sources and storing configuration in structured metadata, Hasura simplifies governance by providing a single source of truth and aligning everyone on what's secure and compliant. This approach also enables faster experimentation and model development with fine-grained permissions for sensitive data, reducing friction and risk associated with traditional data governance methods. With its extensibility, Hasura can convert metadata between different databases, making it easier to manage transitions and reduce duplication of effort. The tool also offers practical use cases such as centralized security and compliance, cross-department data collaboration, real-time policy enforcement, risk management, and audit-ready reporting, ultimately transforming the way financial institutions approach data governance and unlocking new value in their data-driven initiatives.