DataWeaver

DataWeaver is a FastAPI service and Vue UI that sit between Amazon Bedrock AgentCore and business users. Upload tabular data to S3, ask a question, and the service coordinates Bedrock’s code-interpreter tools to deliver Markdown narratives and shareable charts.

Key Features

• Single data ingress – /upload streams files to S3 and returns keyed URIs for analysis sessions.
• Dual hosting model – same container runs inside Bedrock Agent Runtime or in ECS as an API proxy.
• Agent runtime orchestration – LangChain AgentExecutor drives the Bedrock code interpreter to load data, run matplotlib, and save charts per session.
• Polished responses – results return as sanitized Markdown and chart URLs; the Vue front end renders headings, tables, and gallery cards.
• Trace-friendly – headers (traceparent, X-Amzn-Bedrock-AgentCore-Runtime-Session-Id, etc.) and payload overrides are preserved for OTEL.

Architecture

Repository Layout

• server/ – FastAPI backend, Bedrock orchestration utilities, S3 loader, Streamlit helper.
• ui/ – Vue 3 + Vite single page app (axios client, Markdown renderer, chart grid).
• Dockerfile – production image suitable for ECS or Bedrock Agent Runtime. See SETUP.md for detailed installation and deployment steps.

Prerequisites

• Python 3.11+
• Node.js 18+ (for the UI)
• AWS account with Bedrock AgentCore access and S3 permissions
• uv for Python dependency management (optional but recommended)

Backend Setup

# Install uv if needed
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install Python dependencies
uv sync

Create .env in the repository root:

cp .env.example .env

Set the essentials (no static credentials required when using task roles):

AWS_REGION=us-west-2
S3_BUCKET_NAME=your-artifact-bucket
BEDROCK_AGENT_RUNTIME_ARN=arn:aws:bedrock-agentcore:...
CODE_INTERPRETER_TOOL_ID=ci-xxxxxxxx

Run the API locally:

uv run uvicorn server.app:app --host 0.0.0.0 --port 8080

Frontend Setup

cd ui
npm install
npm run dev

Configure the API base URL via .env or runtime global:

• build time: VITE_API_BASE_URL=https://your-api.example.com
• runtime: inject window.__AGENT_CORE_API_BASE_URL

For Markdown playground styling, set VITE_MARKDOWN_PREVIEW=true before starting npm run dev.

API Surface

Method	Path	Description
GET	/ping	Health check
POST	/upload	Multipart file upload → S3 URLs map
POST	/invocations	Bedrock-compatible data analysis endpoint
POST	/chat	Proxy to Bedrock Agent Runtime (UI entry point)

Example /chat request:

curl -X POST "$API_BASE/chat" \
  -H "Content-Type: application/json" \
  -H "X-Amzn-Bedrock-AgentCore-Runtime-Session-Id: session-1234" \
  -d '{
    "s3_urls": {
      "customers": "s3://your-bucket/uploads/customers.csv"
    },
    "prompt": "Generate a bar chart of customer count by country and summarise the top regions."
  }'

Successful responses include:

• output – Markdown narrative
• intermediate_steps – LangChain trace snippets
• dataframes_loaded – DataFrame keys hydrated from S3
• charts – Presigned URLs under s3://bucket/charts/<runtimeSessionId>/

Observability Notes

• ADOT auto-instrumentation (aws-opentelemetry-distro) is installed; supply OTEL variables as needed.
• Trace headers from clients are echoed back under trace.headers and trace.payload.
• The agent prompt instructs Bedrock to avoid leaking session IDs; chart paths remain internal.

Deployment

Docker Build

docker build -t agentcore-dataweaver .

ECS (API proxy)

1. Push image to ECR.
2. Configure task execution role for logging and task role with:
   • bedrock-agentcore:InvokeAgentRuntime
   • S3 read/write to the upload bucket
3. Provide environment variables (BEDROCK_AGENT_RUNTIME_ARN, CODE_INTERPRETER_TOOL_ID, etc.).
4. Expose port 8080 behind your load balancer.

Bedrock Agent Runtime

1. Register the container image with Bedrock Agent Runtime.
2. Supply the same environment variables.
3. Bedrock invokes /invocations and /ping; /chat is unused in this mode.

Available Scripts

• uv run pytest – backend smoke tests (see server/tests).
• npm run dev – UI development server.
• npm run build – production UI bundle.
• npm run lint – hook in your preferred linter.

License

Provided as-is for hackathon experimentation. Adjust, deploy, and extend to fit your environment.