Troubleshooting

When something isn’t working, start here. This page covers the fastest diagnostic checks, then guides you to the right sub-page based on the symptom you’re seeing.

Quick checks

Run through these first — they catch the majority of issues:

1. Bedrock Bedrock
   Amazon Bedrock — a fully managed service for accessing foundation models (embedding and LLM) via a unified API. model access enabled? — Both Titan Embeddings V2 and your LLM must be enabled in the Bedrock console for the deployment region.
2. Correct region? — S3 Vectors, Bedrock models, and inference profiles are not available in all regions. Confirm your terraform.tfvars region matches where you enabled model access.
3. Bootstrap ingest completed? — Check the Lambda Lambda
   AWS Lambda — serverless compute. This project uses two functions: ingest (RSS → embed → store) and query (search → answer). ingest function logs for success. An empty corpus means queries return no results.
4. Cognito Cognito
   Amazon Cognito — a user authentication service providing user pools, JWT tokens, and group-based authorization. user confirmed? — Users must be in CONFIRMED status. An unconfirmed user cannot authenticate to the API.
5. Terraform apply finished cleanly? — Partial applies leave resources in inconsistent states. Check for errors in the apply output.

Tip

Most problems fall into two buckets: missing Bedrock model access (causes AccessDeniedException) or the ingest Lambda not completing (causes empty query results).

Symptom-based navigation

Use the table below to find the right troubleshooting path based on what you’re experiencing.

Symptom	Likely cause	Where to look
Deployment failed (terraform apply errors)	Missing permissions, region not supported, or resource limits	Check Terraform output; see Common Issues
Query returns empty results	Corpus not ingested, time filter too narrow, or vector index empty	Check corpus status below; see Common Issues
AccessDeniedException errors	Bedrock model access not enabled or IAM policy missing permissions	See Bedrock Errors
Web UI not loading or blank page	Amplify Amplify AWS Amplify — a managed hosting service for static web applications. Used to host the briefing UI SPA. deployment incomplete or CORS mismatch	Check Amplify console and CORS config; see Common Issues
Ingest Lambda timeout	Bedrock throttling or slow RSS feeds exceeding 480s limit	See Bedrock Errors (throttling section)
ThrottlingException from Bedrock	Too many concurrent embedding or LLM requests	See Bedrock Errors
401 / 403 from API Gateway API Gateway Amazon API Gateway — a managed HTTP API service with JWT authorization, CORS, and throttling. Routes requests to the Query Lambda.	Expired or invalid JWT token, user not confirmed	Re-authenticate; check Cognito user status

Checking CloudWatch logs

Each Lambda function writes to a dedicated CloudWatch log group. The naming pattern is:

/aws/lambda/<project-prefix>-ingest
/aws/lambda/<project-prefix>-query

To tail logs for the ingest function:

aws logs tail "/aws/lambda/$(terraform output -raw ingest_function_name)" --follow

To tail logs for the query function:

aws logs tail "/aws/lambda/$(terraform output -raw query_function_name)" --follow

Note

Replace the Terraform output commands with your actual function names if you’re not running from the Terraform workspace directory.

Checking corpus status

The query Lambda exposes a /status endpoint through API Gateway API Gateway
Amazon API Gateway — a managed HTTP API service with JWT authorization, CORS, and throttling. Routes requests to the Query Lambda. that returns the current vector count and index health.

# Get the API endpoint
API_URL="$(terraform output -raw api_endpoint)"

# Check status (no auth required for /status)
curl -s "$API_URL/status" | jq .

You can also verify directly that vectors exist in the S3 Vectors S3 Vectors
Amazon S3 Vectors — a purpose-built vector storage capability within S3 that enables similarity search over embeddings without a separate vector database. index:

aws s3vectors list-vectors \
  --vector-bucket-name "$(terraform output -raw vector_bucket_name)" \
  --index-name "$(terraform output -raw vector_index_name)"

If the vector count is zero, the ingest function hasn’t run successfully. Re-trigger it:

aws lambda invoke \
  --function-name "$(terraform output -raw ingest_function_name)" \
  /tmp/ingest-out.json

cat /tmp/ingest-out.json | jq .

Caution

The bootstrap ingest runs automatically during terraform apply via a Lambda Lambda
AWS Lambda — serverless compute. This project uses two functions: ingest (RSS → embed → store) and query (search → answer). invocation resource. If it failed during apply, you’ll need to invoke it manually.

Detailed troubleshooting pages

Bedrock Errors AccessDeniedException, throttling, model validation errors.

Common Issues Deployment failures, empty results, CORS, and timeout problems.