Skip to content

Engineered an agentic retrieval system combining Neo4j (Graph) and Qdrant (Vector) to interpret complex building codes. Automated site reporting by processing construction images with multimodal AI to generate technical logs and status updates.

Notifications You must be signed in to change notification settings

TheDeadcoder/civilmate-backend

Repository files navigation

Setup Guide

Virtual environment

Make sure python3-venv is in your machine

sudo apt install python3-venv

Make a python virtual environment with the following command:

python3 -m venv .venv

activate the environment

source .venv/bin/activate

Install dependencies

Install the required packages with the following command:

pip install -r requirements.txt

Running the backend

To run the backend server, use the following command:

uvicorn app.main:app --reload

The app will start:

http://127.0.0.1:8000/

Plan

  • at first we parse down the entire pdf in batch (we get an array of json each contaning page number and page content)

  • then

  • our graph nodes = our vectors

  • each node is a standalone information point

  • we generate metadata for each node

  • Embed (metadata + node content)

  • we retriece candidates as node's parent

  • either we define a new relationship between 2 newly made nodes

  • or link up with an old one

  • node and vector points share

About

Engineered an agentic retrieval system combining Neo4j (Graph) and Qdrant (Vector) to interpret complex building codes. Automated site reporting by processing construction images with multimodal AI to generate technical logs and status updates.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published