neo4j-labs / llm-graph-builder Goto Github PK

On selection of LLM - user shall be able to process the files on basis of preferred LLM.

1. Add connection modal according to the new Figma
2. Place/Positioning dropzone according to the above design
3. Table css styling fixes

Need to create a common function to call the API's. Some function name changes and logic formatting

can we make all environment variables uppercase.

and add a section to the backend readme on configuration and the env file.

also call out in the file and readme what is optional and what can be overriden e.g. from the client.

It should be more aligned with the usual style of config variables that we use elsewhere:

#OPENAI_API_KEY="sk-..."
DIFFBOT_API_KEY=""
NEO4J_URI=""
NEO4J_USERNAME="neo4j"
NEO4J_PASSWORD=""

I tried to run the app, it still creates duplicates of the file with the same name

and when trying to process the file I get a 422 error

backend   | INFO:     172.18.0.1:44740 - "GET /sources_list HTTP/1.1" 200 OK
backend   | INFO:     172.18.0.1:44732 - "GET /health HTTP/1.1" 200 OK
backend   | INFO:     172.18.0.1:44748 - "GET /health HTTP/1.1" 200 OK
backend   | INFO:     172.18.0.1:44754 - "GET /sources_list HTTP/1.1" 200 OK
backend   | INFO:     172.18.0.1:35078 - "POST /sources HTTP/1.1" 200 OK
backend   | INFO:     172.18.0.1:55262 - "POST /extract HTTP/1.1" 422 Unprocessable Entity

Handle Model Bug fix
Change failed Response Alert position from top center to bottom left
Add a Check for disable state of Generate graph button and dropdown

change the API name from /predict to /extract

• spell out knowledge graph in the description

• rename the body object in the docs to something more consistent and descriptive from Body_kg_creation_predict_post

also add metadata about the file:

• filename
• file-size
• if available file date
• store those a :Source node (or equivalent if the graph transformer already creates a metadata node) in the graph

and in the response at least prepare the numeric processingTime and nodeCount and relationshipCount response fields
and status and errorMessage

Working on handling the bug found while testing
When all the files are in processing then their respective records are populating correctly in table but when lets say processing on on going of 3 files and I upload a new large file with "New" status and doesn't start processing for it then data of records in UI table is getting shuffled .
On refresh retain back to their original data.

1. Fix the white space issue by dynamically adjusting the height and it is responsive even after changing the page size in the table
2. Fix Auto Page Shifting
3. New items should be shown on first page rather than last
4. If the File is Already Processed show it as Completed
5. Removed the extra check for Disabling the Dropdown and Generate Graph Button
6. Connection Modal should display, if user is not connected with Neo4j Database
7. Add the Neo4j Favicon

Also instead of select LLM it should say "Select Model"

Create a setting panel.
Add settings for LLM Dropdown, Access key and Secret Key , Embedding checkbox

When user uploads a file, hit an API [/sources] to post the the file data.

Currently user is able to upload one file at a time. Allow users to add 5 files at a time.

Integrate the relationships created on file upload processing in the table.

As per our understanding the secret key and access key if already available in the source node, put a check of its existence , if its there show the available for the processing/New .

Deploy docker containers to google cloud run , generate a URL

There seems to be a CORS issue.

-> ok seems to be related to the GH codespaces, need to make the backend URL public to make it work for the time being, should be resolved when running it with docker or deploying it elsewhere.

But also connecting to a wrong back-end? Not sure if you hard-coded it, but it should just connect to localhost:8000 on the machine where the backend-is running or the configured base-URL.

Seems you have that hard-coded

https://github.com/neo4j-labs/llm-graph-builder/blob/main/frontend/src/components/DropZone.tsx#L10

1. it should not be hard-coded but come from an .env file (also provide an example.env)
2. it should not just hidden inside a UI component but a proper backend/REST API component !!
3. there should be a health check that valdiates that the backend is correctly running and indicate that to the user!

https://github.com/neo4j-labs/llm-graph-builder/issues/new
Access to XMLHttpRequest at 'https://animated-space-broccoli-jpgjg6pg59qcp7pg-8000.app.github.dev/predict' from origin 'https://studious-dollop-979pxr45x3p4p4-5173.app.github.dev' has been blocked by CORS policy: No 'Access-Control-Allow-Origin' header is present on the requested resource.

POST https://animated-space-broccoli-jpgjg6pg59qcp7pg-8000.app.github.dev/predict net::ERR_FAILED 404 (Not Found)

my backend is running on: https://studious-dollop-979pxr45x3p4p4-8000.app.github.dev/docs

Later: Perhaps mid-term we can even serve them from the backend as static assets.

• rename Source -> Document
• HAS_CHILD relationship inverse (chunk)-[:PART_OF]->(:Document)
• add a single first relationship: (:Document)-[:FIRST_CHUNK]->(:Chunk)
• create a NEXT relationship between chunks of each document

The front end should indicate if the back-end is running.

Right now it shows the file-drop area if neo4j/the backend is connected but there should be a clearer indication.

Add a dropdown for user to select LLM of their choice.

If there is a separate connection information provided in the front-end it should pass that to the backend in a suitable way when making requests.

e.g. for processing files the connection information of the front-end (if available) should be passed on as an extra nested payload and be used in the processing.

Same for listing sources for the table, it should use the front-end connection information.

If the backend is configured with a neo4j connection but the front-end is not connected, it should still work, then automatically using the backends connection config inside the backend.

• by default should come from environment variable / .env file (using dotenv)

• if not set you can use the logic you have here right now, but probably want to add a check and also test for localhost

• https://github.com/neo4j-labs/llm-graph-builder/blob/main/frontend/src/utils/utils.ts#L1

When adding :Source nodes to the graph to represent the files, add a /sources/list endpoint that returns the list of sources ordered by updatedAt descending and returns all the metadata, that was added/updated when creating the nodes

• fileName (for the time being this can be the id - unique constraint)
• fileType
• fileSize
• createdAt
• updatedAt
• processingTime
• status
• errorMessage
• nodeCount
• relationshipCount

Implement a frontend interface element allowing users to specify the S3 bucket name and credential details required for accessing resources.

s3-bucket (with path)

optional credentials (access-key, secret-key) and region (so that public buckets can be accessed without credentials)

https://github.com/neo4j-labs/llm-graph-builder/blob/main/backend/src/main.py#L42-L43

Also add relationship count to the UI

Update the table with the API response.

• instructions how to run / deploy / configure
• link to the public google cloud run URL + link to neo4j workspace
• list of features (upload, s3/gcs, connection to neo4j, file + chunk handling, extract entities with different models, create embeddings, create kNN graph)
• screenshot or short animated gif
• graph model
• screenshot of the graph model in neo4j workspace + query that I shared

Handle validations when user adds invalid url.
Handle state of banner .
Secret and Access key params confirmation.

you have an inconsistentcy on how you use BACKEND_URL -> url() sometimes {url()}sources sometimes {url()}/extract
I changed it now to always use a slash / i.e. {url()}/extract
so that you have to set the environment variable like this without a trailing slash: export BACKEND_API_URL="https://studious-dollop-979pxr45x3p4p4-8000.app.github.dev/"
ideally in url() we would remove trailing slashes

Add some sort of feedback when user clicks on "Generate Graph". The button should show that the files are processing and then indicate completed once the job is done.