AIData Engineering
AI-Powered DBT Assistant
Natural language assistant for dbt — letting non-technical users query the data layer.
Overview
Client Overview
An AI assistant that lets non-technical users perform complex data operations through natural language prompts. Built on Claude via AWS Bedrock, the assistant uses an MCP server to intelligently query the dbt (Data Build Tool) layer, exposed through FastAPI endpoints on AWS EC2.
Industries
AIData Engineering
Technologies
PythonFASTAPIDBTAWS BEDROCK (Claude)AWS EC2
Status
Live & Active
Challenges
The Challenges
1
Translating ambiguous business questions into precise dbt queries.
2
Designing an MCP server that gives Claude rich, structured access to dbt metadata.
3
Running Claude responses through guardrails so untrained users can't break the data layer.
4
Deploying a Python service stack reliably on AWS EC2.
Solutions
Solutions & Strategies
01
MCP-Powered Data Access
- Built an MCP server that exposes dbt models, tests, and metadata to Claude.
- Layered intent recognition so the assistant resolves the right model for each question.
02
AI Orchestration
- Used Claude on AWS Bedrock for high-quality reasoning over data schemas.
- Added retry and validation logic so the assistant returns only verified results.
03
API Surface
- Exposed the assistant through FastAPI endpoints for integration into business tools.
- Deployed on AWS EC2 with a structure ready for scale.
Results
The Results
✓Key Achievements
- Non-technical users querying dbt through natural language in production.
- Claude + MCP integration delivering structured, verified answers.
- FastAPI-exposed assistant integrated into business workflows.
- Reliable AWS EC2 deployment.
★Project Highlights
- Claude (Bedrock) + MCP server architecture.
- Natural language → dbt query translation.
- FastAPI integration surface.
- Guardrailed AI for data safety.
Tech Stack
Technologies Used
AI
AWS BedrockClaudeMCP
Backend
PythonFastAPI
Data
dbt
Infrastructure
AWS EC2
