Intelligent Document Query Platform — GitHub-ready Low-Level Design (LLD)

This repository contains a ready-to-use Low-Level Design (LLD) for the Intelligent Document Query Platform. It is organized so you can drop each file into a GitHub repo and iterate from there.

---

Repository layout (suggested)

intelligent-doc-query-lld/
├─ README.md                # High-level summary + how to use this LLD
├─ ARCHITECTURE.md          # Detailed architecture diagrams (mermaid + explanation)
├─ docs/
│  ├─ sequence_diagrams.md  # Mermaid diagrams for sequences
│  └─ flow_notes.md         # Additional flow explanations
├─ terraform/
│  ├─ main.tf
│  ├─ variables.tf
│  ├─ outputs.tf
│  └─ modules/
├─ backend/
│  ├─ lambda_handlers/
│  │  ├─ presign_upload.py
│  │  ├─ document_processor.py
│  │  └─ ask_question.py
│  ├─ requirements.txt
│  └─ graphql/
│     └─ schema.graphql
├─ frontend/
│  ├─ README_FRONTEND.md
│  ├─ src/
│  │  ├─ App.tsx
│  │  ├─ components/
│  │  └─ graphql/
│  │     └─ queries.ts
│  └─ package.json
├─ db/
│  ├─ schema.sql
│  └─ migrations/
├─ .github/
│  └─ workflows/
│     └─ deploy.yml
└─ LLD.md                   # This low-level design in markdown (detailed)

---

LLD

1. Overview

This repo contains the Low-Level Design (LLD) for the Intelligent Document Query Platform. The design is cloud-native, serverless, and event-driven. The main components are:

• React+TypeScript SPA frontend
• Serverless Python backend (AWS Lambda) behind API Gateway (GraphQL)
• PostgreSQL (RDS) with pgvector for vector similarity search
• S3 for document storage
• Terraform for IaC
• GitHub Actions for CI/CD

---

2. Architecture (ARCHITECTURE.md)

Include this mermaid diagram in ARCHITECTURE.md or your README to render visually on GitHub:

flowchart LR
  subgraph User
    U[User browser]
  end
  subgraph CDN
    CF[CloudFront] --> S3Static[S3 (React static files)]
  end
  U --> CF
  U --> APIGW[API Gateway (GraphQL)]
  APIGW --> LAMBDA[Lambda functions]
  LAMBDA -->|reads/writes| RDS[(Postgres + pgvector)]
  LAMBDA -->|reads/writes| S3[Documents bucket]
  LAMBDA --> LLM[LLM Provider API]
  S3 -->|Event| ProcessingLambda[Processing Lambda]
  ProcessingLambda --> RDS

Add a short explanation of each box and security considerations (VPC, IAM roles, KMS for secrets).

---

3. Database Schema (db/schema.sql)

-- PostgreSQL schema for documents and chunks
CREATE EXTENSION IF NOT EXISTS pgcrypto;
CREATE EXTENSION IF NOT EXISTS vector; -- pgvector

CREATE TABLE IF NOT EXISTS documents (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  user_id VARCHAR(255),
  file_name VARCHAR(1024),
  s3_key VARCHAR(2048) UNIQUE,
  status VARCHAR(32) CHECK (status IN ('UPLOADED','PROCESSING','READY','ERROR')) DEFAULT 'UPLOADED',
  created_at TIMESTAMP WITH TIME ZONE DEFAULT now(),
  updated_at TIMESTAMP WITH TIME ZONE DEFAULT now()
);

CREATE TABLE IF NOT EXISTS document_chunks (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  document_id UUID REFERENCES documents(id) ON DELETE CASCADE,
  chunk_text TEXT,
  embedding vector(384),
  created_at TIMESTAMP WITH TIME ZONE DEFAULT now()
);

-- Example index for ivfflat (tune lists depending on dataset)
-- CREATE INDEX document_chunks_embedding_idx ON document_chunks USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);

-- Optional table for sessions / metadata
CREATE TABLE IF NOT EXISTS sessions (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  user_id VARCHAR(255),
  metadata JSONB,
  created_at TIMESTAMP WITH TIME ZONE DEFAULT now()
);

Notes: vector(384) dimension should match the embedding model used (e.g., some models use 384, 512, 1536, etc.).

---

4. GraphQL schema (backend/graphql/schema.graphql)

type PresignedUrlResponse {
  uploadUrl: String!
  s3Key: String!
}

type Document {
  id: ID!
  fileName: String!
  status: String!
}

type Answer {
  text: String!
  sources: [String!]
}

type Mutation {
  generatePresignedUploadUrl(fileName: String!, fileType: String!): PresignedUrlResponse!
  askQuestion(documentId: ID!, question: String!): Answer!
}

type Query {
  getDocumentStatus(documentId: ID!): Document
}

# Subscription (note: serverless subscriptions may use Websockets or AppSync)
type Subscription {
  answerStream(documentId: ID!, question: String!): String
}

---

5. Backend skeleton (Python Lambda handlers)

backend/lambda_handlers/presign_upload.py

import os
import boto3
import json

s3 = boto3.client('s3')
BUCKET = os.environ.get('DOC_BUCKET')


def handler(event, context):
    body = json.loads(event.get('body','{}'))
    file_name = body['fileName']
    file_type = body.get('fileType','application/octet-stream')

    s3_key = f"uploads/{file_name}"
    presigned = s3.generate_presigned_url(
        'put_object',
        Params={'Bucket': BUCKET, 'Key': s3_key, 'ContentType': file_type},
        ExpiresIn=900
    )

    response = {'uploadUrl': presigned, 's3Key': s3_key}
    return {
        'statusCode': 200,
        'body': json.dumps(response),
        'headers': {'Content-Type': 'application/json'}
    }

backend/lambda_handlers/document_processor.py

# This lambda is triggered by S3 put events
import os
import boto3
import psycopg2

# Pseudocode: download, extract text, chunk, embed, insert into RDS

def handler(event, context):
    # 1. Get S3 key
    # 2. Download file
    # 3. Extract text (use textract, pdfminer, etc.)
    # 4. Chunk text (e.g., 500 tokens or 1000 chars)
    # 5. Create embeddings (sentence-transformers or remote provider)
    # 6. Insert into document_chunks table
    # 7. Update documents.status = 'READY'
    return {'status': 'ok'}

backend/lambda_handlers/ask_question.py

# Handles the askQuestion GraphQL mutation
# 1. Generate embedding for the question
# 2. Query pgvector top-k similar chunks
# 3. Build prompt + call LLM (streaming if supported)
# 4. Return Answer object (and optionally push stream messages)

Add requirements.txt with libraries like boto3, psycopg2-binary, requests, sentence-transformers (if used), and any GraphQL handler library you prefer.

---

6. Frontend skeleton

frontend/src/App.tsx (TypeScript + React sketch)

import React from 'react'
import { ApolloClient, InMemoryCache, ApolloProvider } from '@apollo/client'

const client = new ApolloClient({
  uri: process.env.REACT_APP_GRAPHQL_URL,
  cache: new InMemoryCache(),
})

export default function App(){
  return (
    <ApolloProvider client={client}>
      <div className="container mx-auto p-4">
        <h1 className="text-2xl">Intelligent Document Query</h1>
        {/* Upload flow and Ask widget components go here */}
      </div>
    </ApolloProvider>
  )
}

Provide frontend/README_FRONTEND.md with instructions to run (npm install, npm start) and environment variables like REACT_APP_GRAPHQL_URL.

---

7. Sequence diagrams (docs/sequence_diagrams.md)

Add the two sequence diagrams from the LLD in mermaid format so GitHub can render them. Example (already provided in your original LLD):

sequenceDiagram
    participant User
    participant Frontend
    participant API
    participant S3
    participant ProcessingLambda

    User->>Frontend: Selects file
    Frontend->>API: generatePresignedUploadUrl()
    API->>S3: generate presigned URL
    Frontend->>S3: upload using URL
    S3->>ProcessingLambda: S3 event
    ProcessingLambda->>RDS: store chunks and embeddings

---

8. Terraform (terraform/main.tf) — minimal example

provider "aws" {
  region = var.region
}

resource "aws_s3_bucket" "static_site" {
  bucket = var.static_bucket_name
  acl    = "private"
}

resource "aws_iam_role" "lambda_exec" {
  name = "lambda_exec_role"
  assume_role_policy = data.aws_iam_policy_document.lambda_assume.json
}

# More resources: api gateway, lambda functions, rds, cloudfront, vpc

Add variables.tf and outputs.tf and split complex resources into modules/ for maintainability.

---

9. GitHub Actions (.github/workflows/deploy.yml)

name: CI/CD
on:
  push:
    branches: [ main ]

jobs:
  build-and-test-backend:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Set up Python
        uses: actions/setup-python@v4
        with:
          python-version: '3.11'
      - name: Install dependencies
        run: |
          pip install -r backend/requirements.txt
      - name: Run backend tests
        run: |
          cd backend && pytest || true

  build-and-test-frontend:
    runs-on: ubuntu-latest
    needs: build-and-test-backend
    steps:
      - uses: actions/checkout@v4
      - name: Setup Node
        uses: actions/setup-node@v4
        with:
          node-version: '20'
      - name: Install & Test
        run: |
          cd frontend
          npm ci
          npm run build --if-present

  deploy-infrastructure:
    runs-on: ubuntu-latest
    needs: build-and-test-frontend
    steps:
      - uses: actions/checkout@v4
      - name: Setup Terraform
        uses: hashicorp/setup-terraform@v2
      - name: Terraform init & apply
        env:
          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
        run: |
          cd terraform
          terraform init
          terraform apply -auto-approve

  deploy-application:
    runs-on: ubuntu-latest
    needs: deploy-infrastructure
    steps:
      - uses: actions/checkout@v4
      - name: Deploy frontend to S3
        run: |
          aws s3 sync frontend/build s3://${{ secrets.STATIC_BUCKET_NAME }} --delete
        env:
          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}

      - name: Deploy Lambdas
        run: |
          echo "Package and deploy lambda functions (use terraform or aws cli / sam)")

Adjust tests and packaging commands to match your project specifics.

---

10. Security & Operations Notes

• Use IAM least privilege for Lambda & processing roles.
• Store secrets in AWS Secrets Manager / Parameter Store and fetch at runtime.
• Use CloudWatch Logs + structured logging for observability.
• Use VPC and security groups for RDS. Ensure Lambda has required access (either via VPC or through RDS proxy).
• Use KMS for encrypting S3 and RDS snapshots as needed.

---

11. Implementation checklist (tasks to convert LLD -> code)

1. Create GitHub repo and push this layout.
2. Implement Terraform modules for core infra.
3. Implement Lambda functions with packaging (SAM or Terraform zip deploy).
4. Configure API Gateway (GraphQL) and subscriptions (websocket or AppSync).
5. Implement frontend upload + ask flows with GraphQL calls.
6. Test full pipeline with a small PDF and an LLM provider.

---

12. CHANGELOG / CONTRIBUTING

Add a CONTRIBUTING.md and CODE_OF_CONDUCT.md if you expect external collaborators.

---

Appendix: Quick reference snippets

SQL: ivfflat index creation

-- Requires pgvector >= x.x
CREATE INDEX IF NOT EXISTS idx_doc_chunks_embedding ON document_chunks USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);

Mermaid: Ask question sequence

sequenceDiagram
    participant User
    participant Frontend
    participant API
    participant RDS
    participant LLM

    User->>Frontend: Ask question
    Frontend->>API: askQuestion()
    API->>RDS: find top-k chunks (pgvector)
    RDS-->>API: returns chunks
    API->>LLM: send prompt
    LLM-->>API: stream response
    API-->>Frontend: forward stream

---

End of LLD