Job Profiles Vector Store

A Python application that creates and queries a vector store of job profiles using LangChain, ChromaDB, and HuggingFace embeddings for semantic search capabilities.

Features

• Processes CSV files containing job profiles
• Creates embeddings using HuggingFace's all-MiniLM-L6-v2 model
• Implements efficient batch processing for large datasets
• Provides similarity search functionality for job profiles
• Implements daily token limiting for Azure API calls

Prerequisites

• Python 3.10+
• Poetry for dependency management

Installation

1. Clone the repository
2. Install dependencies using Poetry:

poetry install

3. Register for a LLM API, such as Mistral Small through Azure Foundry
4. Rename .env.sample into sample and set your environment variables
5. Copy data into the root folder so you have /data/job profiles/2025-02-07_profiles.csv

Token Limiting

The application includes a token limiting system to prevent excessive usage of the Azure API:

• Set the daily token limit using the AZURE_DAILY_TOKEN_LIMIT environment variable (default: 100,000)
• Token usage is tracked and reset daily
• When the limit is reached, API calls will return a 429 error
• View current token usage statistics at the /token-usage endpoint

Token Usage Endpoint

GET /token-usage

Response:

{
  "daily_usage": 5000,
  "daily_limit": 100000,
  "total_usage": 25000,
  "remaining_tokens": 95000,
  "last_reset_date": "2023-06-01"
}

Running Open WebUI with local Kubernetes

Follow these instructions to run a UI that allows you to run queries with the RAG pipeline. Before getting started, ensure you have job_profiles_db2 in the root by building the vector store through the create_vectore_store.ipynb notebook (see below).

Build and run API docker image

If dependencies changed, generate requirements.txt from poetry and copy to /backend: poetry export --format=requirements.txt --output requirements.txt --without-hashes

Build docker image (run from root):

docker build -f backend/Dockerfile -t rag-backend:local .

To run the api docker image:

docker run --env-file ./.env -p 8000:8000 rag-backend:local

Setup Kubernetes

Select local context: kubectl config use-context docker-desktop

To view contexts: kubectl config get-contexts

Add nginx ingress controller: kubectl apply -f https://raw.githubusercontent.com/kubernetes/ingress-nginx/controller-v1.12.0/deploy/static/provider/cloud/deploy.yaml

Check that ingress controller is working: kubectl get pods -n ingress-nginx

Create namespace: kubectl create namespace fd34fb-dev

Apply local overlay configuration: kubectl apply -k kubernetes/openwebui/overlays/local

Open WebUI should be available at (ignore security warning): https://kubernetes.docker.internal/

Go to Admin Panel -> Settings -> Connections, delete all records. Under OpenAI API add a new record with http://external-api:8000/v1 for URL and anything for the Key

The model should now be available on the main selection screen and querying should work.

Install kubernetes dashboard

kubectl apply -f https://raw.githubusercontent.com/kubernetes/dashboard/v2.5.1/aio/deploy/recommended.yaml

kubectl create serviceaccount admin-user -n kubernetes-dashboard

kubectl create clusterrolebinding admin-user --clusterrole=cluster-admin --serviceaccount=kubernetes-dashboard:admin-user

kubectl -n kubernetes-dashboard create token admin-user

kubectl proxy

Go to: http://localhost:8001/api/v1/namespaces/kubernetes-dashboard/services/https:kubernetes-dashboard:/proxy/

Notebooks Usage

The project contains two Jupyter notebooks in the notebooks directory:

1. create_vectore_store.ipynb: Creates the vector store from job profile data
2. try_vector_store.ipynb: Demonstrates how to query the vector store

To run notebooks in WSL

Update python:

sudo apt update && sudo apt upgrade

sudo apt upgrade python3

Install poetry:

curl -sSL https://install.python-poetry.org | python3 -

echo 'export PATH="/home/<USER_NAME>/.local/bin:$PATH"' >> ~/.bashrc

source ~/.bashrc

Add vs code to path in .bashrc:

export PATH="/mnt/c/Users/<USER_NAME>/AppData/Local/Programs/Microsoft VS Code/bin:$PATH"

If you get /usr/bin/env: 'sh': No such file or directory, ensure you have this in .bashrc:

export PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:$PATH

Then navigate to ex.:

cd /mnt/c/Users/<USER_NAME>/GitHub/job-profiles-rag

Start vs code:

code .

Install WSL extension for vs code and restart vs code.

Install Python extension (should show up under WSL: UBUNTU - INSTALLED)

Run poetry install

If you get [Errno 2] No such file or directory: 'python, you may need to create a symlink from python3 to python:

sudo ln -s /usr/bin/python3 /usr/bin/python

Get poetry environment path:

poetry env info --path

Open VS Code Command Palette (Ctrl+Shift+P) and:

• Select "Python: Select Interpreter"
• Choose "Enter interpreter path"
• Paste the Poetry environment path + "/bin/python", e.g. /home/<USER_NAME>/.cache/pypoetry/virtualenvs/job-profiles-rag-gqqwjc62-py3.10/bin/python

Install kernel for the poetry environment:

poetry run ipython kernel install --name "job-profiles-rag" --user

Then in jupyter notebook:

• Click the kernel picker in the top right of the notebook
• Choose the kernel matching your Poetry environment

Add data folder to the root directory, so you have /data/job profiles/2025-02-07_profiles.csv

Docker - other

Open WebUI in docker

To run open webui in docker such that it can connect to api running on localhost:

docker run -d -p 3000:8080 \
  --add-host=host.docker.internal:host-gateway \
  -e OPENAI_API_BASE_URL=http://host.docker.internal:8000 \
  -e ENABLE_OAUTH_SIGNUP="true" \
  -e OAUTH_CLIENT_ID="xxxxx" \
  -e OAUTH_CLIENT_SECRET="xxxxx" \
  -e OPENID_PROVIDER_URL="https://dev.loginproxy.gov.bc.ca/auth/realms/standard/.well-known/openid-configuration" \
  -e OAUTH_PROVIDER_NAME="Keycloak" \
  -e OAUTH_AUTO_SIGNUP="true" \
  -e OAUTH_MERGE_ACCOUNTS_BY_EMAIL="true" \
  -v open-webui:/app/backend/data \
  --name open-webui \
  --restart always \
  ghcr.io/open-webui/open-webui:main

In open webui set api url to:

http://host.docker.internal:8000/v1

To update local openwebui image:

docker pull ghcr.io/open-webui/open-webui:main

Stop and remove the current container docker rm -f open-webui

License

This project is licensed under the terms included in the LICENSE file.

Show current packages vs latest versions

poetry show --latest

Show debug code for langchan processes

from langchain.globals import set_debug
set_debug(True)

Debug everything

import logging logging.basicConfig(level=logging.DEBUG)

Add a shortcut to execute in the current Debug Console:

Open keyboard shortcuts: Ctrl+K+S. Search for Debug: Evaluate in Debug Console and select a shortcut, like Alt+Shift+E

To set custom graphics

docker cp backend/assets/splash.png open-webui:/app/backend/open_webui/static/splash.png

docker cp backend/assets/square/android-chrome-192x192.png open-webui:/app/backend/open_webui/static/favicon.png

docker cp backend/assets/square/android-chrome-512x512.png open-webui:/app/backend/open_webui/static/logo.png

docker cp backend/assets/square/favicon.ico open-webui:/app/build/favicon/favicon.ico

docker cp backend/assets/square/android-chrome-192x192.png open-webui:/app/build/favicon/favicon-96x96.png

To run openwebui with built-in ollama service:

docker run -d -p 3000:8080 --gpus all -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui-gpu --restart always ghcr.io/open-webui/open-webui:ollama

Data will be contained in /root/.ollama/app/backend/data

Restart ollama:

taskkill /F /IM "ollama app.exe" /IM ollama.exe ollama serve

ollama list: See models ollama ps: Displays currently running models ollama stop: Stops a running model ollama rm: Removes a model from the system

To run a new model from HF:

ollama run hf.co/username/repository

Enabled mirrored mode in WSL (allows treating localhost to connect to services running on windows):

Add .wslconfig in C:\Users\user_name with:

[wsl2]
networkingMode=mirrored

To run neo4j in docker:

(just run docker-compose up in /neo-4j instead)

docker run \
    --name neo4j-db \
    --publish=7474:7474 --publish=7687:7687 \
    --volume=$HOME/neo4j/data:/data \
    --volume=$HOME/neo4j/logs:/logs \
    --env NEO4J_AUTH=neo4j/your_password \
    --restart always \
    neo4j:latest

Then login here: http://localhost:7474/

Enable wrapping in jupyter notebooks in vscode:

Open VS Code Settings (⌘ + , on Mac / Ctrl + , on Windows).

Search for "Notebook > Output: Word Wrap".

Check the box to enable this setting or add to settings.json:

"notebook.output.wordWrap": true

To view all keys in neo4j (useful to find which fields to embed)

MATCH (d:Document) 
RETURN keys(d) LIMIT 1

To dump neo4j db (ensure db is stopped):

docker run --rm \
  --user 7474:7474 \
  -v $HOME/neo4j/data:/data \
  -v $HOME/neo4j/dumps:/dumps \
  neo4j:5.26.2 \
  neo4j-admin database dump neo4j --to-path=/dumps

To restore from new4j database dump (ensure db is stopped):

sudo chmod -R 777 neo4j

docker run --rm \
  --user 7474:7474 \
  -v $HOME/neo4j/data:/data \
  -v $HOME/neo4j/dumps:/dumps \
  neo4j:5.26.2 \
  neo4j-admin database load neo4j --from-path=/dumps --overwrite-destination=true

See node labels and property distribution in the database

MATCH (n) 
RETURN labels(n) AS nodeLabels, 
     keys(n) AS properties, 
     count(*) AS frequency
ORDER BY frequency DESC

To query unstructured data:

MATCH (node)
WITH node, 1.0 as score
OPTIONAL MATCH (node)-[r]->(related)
RETURN coalesce(node.text, node.name) AS text,
       score,
       node {.*, labels: labels(node)} AS metadata,
       collect({relationship: type(r), node: related}) AS graph_context

For heterogeneous graphs, assign a common secondary label to all nodes while preserving original labels:

MATCH (n) SET n:Entity