SS

Search

Jump to a page, project or post

All projects
Shipped

Personal RAG assistant on AWS

A retrieval-augmented chat service built to learn the full build-and-deploy loop: NestJS, MongoDB Atlas vector search, Hugging Face embeddings, Groq-hosted Llama, Docker, and a hand-provisioned EC2 box.

Role
Everything
Date
Kind
Personal
Personal RAG assistant on AWS: system sketchingestDocumentschunkedQuestionNestJS serviceDocker on EC2Hugging FaceembeddingsAtlas vector searchMongoDBLlama via Groqanswers
System sketch. Boxes are the real components; packets show the direction of flow.
Retrieval
Atlas Vector Search
LLM
Llama via Groq
Deploy
Docker on EC2, ap-south-1
  • NestJS
  • TypeScript
  • MongoDB Atlas
  • Hugging Face
  • Groq
  • Docker
  • AWS EC2
GitHub

Why

I wanted to own every step from git init to a public URL, including the parts a platform usually hides: security groups, SSH, running a container on a server, and what breaks when the instance restarts.

What it does

  • Ingests documents, chunks them, embeds them with a Hugging Face model and stores vectors in MongoDB Atlas.
  • Answers questions over that corpus with a Llama model served by Groq.
  • Runs as a single Docker container on an EC2 instance.

Loose ends I know about

No Elastic IP yet, no domain or TLS, and the deploy is a manual pull. Fixing those is part of the cloud and DevOps roadmap.