Migrating an Agentic RAG App to AWS Serverless
A previous post covered building an agentic RAG router over **LLM Zoomcamp * capstone submissions. This series is about moving it off a single EC2 box and onto AWS serverless, where nearly every obvious choice turned out to be wrong, and the most valuable decision was deleting a database rather than migrating it. This part covers the starting point and the resulting architecture; The course on…
The previous article introduced an agentic RAG router for processing LLM theses submitted at Zoomcamp. This article focuses on migrating the project from a single EC2 server to a serverless AWS environment. The starting point and resulting architecture will be examined, beginning with the current setup and its drawbacks.
Currently, the assistant operates on a single t4g.small EC2 instance, running six Docker Compose containers. This setup uses RAM for loading and rebuilding a corpus of 24,775 fragments, totaling 115MB of JSON. Two local machine learning models require a large virtual environment and an image. The system lacks infrastructure as code and deployment is done via a bash script, resulting in a monthly cost of about $15 regardless of usage.
To improve the system, it will be rebuilt as a fully serverless AWS application using real infrastructure as code (IaC). The development will utilize Amazon's resources, such as Nova for output, Titan for embeddings, Amazon Rerank for reranking, and boto3—the only SDK used. OpenAI and Anthropic will not be used.
The new architecture consists of several main components: CloudFront for CDN and edge origin checks, S3 for static SPA files, Lambda for the API, Bedrock for embeddings, vector search, and Reranking, S3 Vectors for storing 25,482 vectors, S3 artifacts for projects, SQLite for projects, chunks, and FTS5, DynamoDB for conversations, feedback, and budget calculations, and Step Functions for scraping and mapping tasks.
All components will leverage serverless AWS services, ensuring no resources are idling and the corpus is no longer a runtime cost.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
