document retrieval caching

document-retrieval

Build a backend for document retrieval with caching.

Caching Algorithm Created a Queue in server which will store limited number of nodes from vector database from previous quries. For every query from user will be first matched with queries in queue and if similarity (cosine) is above 0.9 than those nodes returned back from the lookup table maintained seperately.

Results

First request => time taken 787 ms

Second Similar request => time taken 54 ms

Bad request

Example Request

---

Health check endpoint

curl --location 'localhost:8001/health'

Search check endpoint

curl --location 'localhost:8001/search' --header 'user_id: riondsilva' --header 'Content-Type: application/json' --data '{ "text": "Ingredient Substitutions" , "top_k": 5 , "threshold": 0.3 }'

Docker Image

---

*Run in order

Chroma Image

docker run -p 8000:8000 chromadb/chroma

Postgres Image

docker run --name user-database -p 5432:5432
 -e POSTGRES_PASSWORD=your_password -d postgres:12.20-alpine3.20   

docker stop user-database

Doc-Retrieval

git clone https://github.com/RionDsilvaCS/document-retrieval-caching.git

Create .env file in the main directory and add the below variables

DATABASE_URL="postgresql://postgres:your_password@localhost:5432/postgres"
COLLECTION_NAME="collection_name"

Create directory name data and add pdf documents

mkdir data

Run store_in_db.py to save the ./data content to chromadb

python store_in_db.py

docker compose up --build

API Endpoints

---

/health is GET endpoint to check if server is boot up

Endpoint

http://127.0.0.1:8001/health

Response

{
  "Server Status": "Successfully running ?"
}

/search is a POST endpoint to conversate with the chatbot

Endpoint

http://127.0.0.1:8001/search

Header

{
  "user_id": "riondsilva"
}

JSON Body

{
  "text": "How are you?",
  "top_k": 3,
  "threshold": 0.3
}

Response

{
  "bot_message": {
    "top_k_results": [
      {
        "message": "I am great ?"
      },
      {
        "message": "I am fine"
      },
      {
        "message": "I am good"
      }
    ],
    "top_k": 3,
    "threshold": 0.9
  },
  "user_message": {
    "text": "How are you?",
    "top_k": 3,
    "threshold": 0.9,
    "inference_time (secs)": 1.16
  },
  "user_info": "riondsilva"
}