ABC Atlas Assistant Technical Documentation

Overview

Release version 1.0.0

The ABC Atlas Assistant is an AI-driven assistant that enables users to interact with ABC Atlas datasets and metadata using natural language.

This release establishes the foundation for conversational data exploration across the ABC Atlas.

Provider: AWS Bedrock

Model: Claude Haiku 4.5

Leveraging: Cache Prompting, Vercel AI SDK

Agent Tools

Client Tools

Server Tools

Agent Capabilities

Natural Language Prompting

  • Search across ABC Atlas datasets and visualizations

  • Interaction with the ABC Atlas user guide

  • Intelligent mapping of user intent to relevant datasets and APIs via agent tools

Visualization Capabilities

  • Configure visualization parameters including filtering and coloring by cell properties and genes, plus dataset searching and selection

  • Native rendering using platform components

Contextual Awareness

  • Conversation history of four messages is maintained for follow-up queries

  • Context-aware responses that understand previous interactions

  • Enriched answers through metadata integration

Platform Integration

  • Leverages existing Platform APIs

  • Retrieval-Augmented Generation (RAG) for accurate, grounded responses

  • Modular back-end architecture supporting future data source expansion

Technical Improvements

Back-end Architecture

  • GraphQL resolvers with dynamic schema expansion for conversation history logging

  • Streaming responses for improved user experience

Performance & Reliability

  • Optimized API handling with asynchronous processing

  • Enhanced caching for system prompts and tool calls

  • Robust error handling with fallback responses